UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Genetics of adaptation in experimental populations of yeast Ono, Jasmine 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2018_february_ono_jasmine.pdf [ 7.87MB ]
Metadata
JSON: 24-1.0362398.json
JSON-LD: 24-1.0362398-ld.json
RDF/XML (Pretty): 24-1.0362398-rdf.xml
RDF/JSON: 24-1.0362398-rdf.json
Turtle: 24-1.0362398-turtle.txt
N-Triples: 24-1.0362398-rdf-ntriples.txt
Original Record: 24-1.0362398-source.json
Full Text
24-1.0362398-fulltext.txt
Citation
24-1.0362398.ris

Full Text

Genetics of adaptation in experimentalpopulations of yeastbyJasmine OnoB.Sc., The University of Toronto, 2008M.Sc., The University of British Columbia, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Zoology)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)December 2017© Jasmine Ono 2017AbstractEvolution proceeds through genetic changes to individuals, which are either propagated or disappear overgenerations. Adaptation is one of the main mechanisms driving these changes in genetic composition. Spe-ciation can also result from different, and incompatible, genetic changes occurring in different populations.This thesis furthers our knowledge of the genetics of adaptation and speciation using the budding yeast Sac-charomyces cerevisiae. My work on the genetic basis of adaptation to high concentrations of copper, whencontrasted with a similar experiment using the fungicide nystatin, showed that the environment has a stronginfluence on both the number of genes that are the targets of selection and the types of potentially beneficialmutations. These results have implications for the repeatability of genetic evolution. In a second study, Ifound that genetic interactions between individually isolated single-step beneficial mutations from the sameselective environment often exhibited the type of epistasis that underlies speciation even though these mu-tations occurred within a single biosynthetic pathway. These results support the mutation-order model ofspeciation by adaptation, where the chance order of mutations in separated populations leads to divergenceand the build-up of reproductive isolation due to genetic incompatibility. Negative genetic interactions be-came positive when the level of stress was increased, indicating that genetically-based reproductive isolationcan also be environment-dependent. Finally, I found that diploid yeast were generally not able to adapt toa level of fungicide to which haploid yeast can adapt. Diploids have been found to adapt to a lower con-centration of the same drug, indicating that the exact environment (type and concentration) and ploidy canhave an impact on the likelihood of genetic rescue. Together, these results have implications for our under-standing of the genetic basis of adaptation in different types of environments and different levels of the sameenvironmental stressor.iiLay SummaryMany aspects of an organism are encoded in their DNA, including their ability to tolerate stressful envi-ronments. Changes to DNA can therefore change how well suited an organism is to certain conditions, andbetter-adapted types have an advantage over others. Throughout this thesis I have studied such changes usingexperimentally-evolved populations of the common brewing and baking yeast, Saccharomyces cerevisiae. Indoing so, I have found that the process of adaptation, and the underlying changes involved, can depend ona variety of factors including the nature of the environment, the level of stress imposed and the genome ofthe organism. In addition, I find that different genetic solutions to the same adaptive problem are not alwayscompatible with each other. This incompatibility can lead to the evolution of new species.iiiPrefaceA version of Chapter 2 has been published as “Gerstein, A. C., Ono, J., Lo, D. S., Campbell, M. L., Kuzmin,A., & Otto, S. P. (2015). Too much of a good thing: the unique and repeated paths toward copper adaptation.Genetics, 199(2), 555-571” with A.C. Gerstein and I listed as co-first authors. A.C. Gerstein and S.P. Ottoconceived the original project, and I conceived and carried out the genetic analyses and subsequent assaysof the lines to determine which mutations caused adaptation. I performed laboratory work in conjunctionwith A.C. Gerstein, D.S. Lo, M.L. Campbell and A. Kuzmin. S.P. Otto wrote the scripts to analyze theIllumina sequence data and determined the expected frequency of mutations causing nonsynonymous andstop codons. I performed analyses of copper tolerance of deletion lines and of single mutations and A.C.Gerstein performed other phenotypic analyses, with advice from S.P. Otto. The manuscript was written incollaboration with A.C. Gerstein and S.P. Otto. I wrote the initial draft of the sections on single mutationsand the discussion and A.C. Gerstein, S.P. Otto and I all contributed major revisions to the manuscript.A version of Chapter 3 has been published as “Ono, J., Gerstein, A. C., & Otto, S. P. (2017). Widespreadgenetic incompatibilities between first-step mutations during parallel adaptation of Saccharomyces cerevisiaeto a common environment. PLoS biology, 15(1), e1002591”. I conceived the project in conjunction withA.C. Gerstein and S.P. Otto. I performed the laboratory work and performed the analyses with advice fromA.C. Gerstein and S.P. Otto. I prepared the majority of the visualizations, although Fig B.4 was prepared byA.C. Gerstein. I prepared the majority of the initial draft, with assistance from A.C. Gerstein and S.P. Otto.We all contributed major revisions to the manuscript.A version of Chapter 4 is in preparation for publication in collaboration with A. Kuzmin, L. Miller andS.P. Otto. S.P. Otto and I conceived the original project. I performed laboratory work in conjunction with A.Kuzmin and L. Miller. S.P. Otto wrote the code to determine the expected number of mutations per site. Iperformed all analyses and prepared the initial draft of the manuscript, and S.P. Otto contributed revisions tothe manuscript.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Genetic model organisms and the study of evolution . . . . . . . . . . . . . . . . . . . . . 21.2 Repeatability of evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Genetic interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.3.1 Bateson-Dobzhansky-Muller model of speciation . . . . . . . . . . . . . . . . . . . 61.4 Limits to adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71.5 What can large-scale datasets tell us about evolution? . . . . . . . . . . . . . . . . . . . . . 81.5.1 How well do mutation collections represent natural mutations? . . . . . . . . . . . 91.5.2 How consistent are genetic interactions across levels of biological diversity? . . . . 131.5.3 How do genetic interactions depend on the environment? . . . . . . . . . . . . . . 181.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 Too much of a good thing: The unique and repeated paths toward copper adaptation . . . . 212.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.2.1 Evolution of haploid mutation lines . . . . . . . . . . . . . . . . . . . . . . . . . . 232.2.2 Sequencing of haploid mutation lines . . . . . . . . . . . . . . . . . . . . . . . . . 272.2.3 Expected frequency of mutations causing non-synonymous and stop codons . . . . 272.2.4 Determination of CUP1 copy number . . . . . . . . . . . . . . . . . . . . . . . . . 28vTable of Contents2.2.5 Phenotypic assays of CBM lines . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.2.6 Copper tolerance of deletion lines . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.2.7 Tetrad dissections to isolate single mutations . . . . . . . . . . . . . . . . . . . . . 302.2.8 Fitness effect of single mutations on growth rate and copper tolerance . . . . . . . . 312.2.9 Data Accessibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.3.1 Single base-pair changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.3.2 Aneuploidies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.3.3 Mutagenic effects of copper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.3.4 Phenotypic assays of CBM lines . . . . . . . . . . . . . . . . . . . . . . . . . . . 372.3.5 Tetrad dissections to isolate single mutations . . . . . . . . . . . . . . . . . . . . . 402.3.6 Reexamining the petite mutations . . . . . . . . . . . . . . . . . . . . . . . . . . . 422.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423 Widespread Genetic Incompatibilities Between First-Step Mutations During Parallel Adapta-tion of Saccharomyces cerevisiae to a Common Environment . . . . . . . . . . . . . . . . . . 463.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463.1.1 Epistasis and its role in evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 463.1.2 Determinants of epistasis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.1.3 Reproductive incompatibilities in nature and in the lab . . . . . . . . . . . . . . . . 483.1.4 Investigation of epistasis between first-step mutations . . . . . . . . . . . . . . . . 493.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.2.1 Strain construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.2.2 Growth rate assays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.2.3 Tolerance across a range of nystatin . . . . . . . . . . . . . . . . . . . . . . . . . . 533.2.4 Sterol Assay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.2.5 Outlier detection and removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.2.6 Statistical analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.3.1 Epistasis of haploids in nystatin . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.3.2 Comparison of epistasis between haploids and diploids . . . . . . . . . . . . . . . . 563.3.3 Epistasis for growth in YPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.3.4 Tolerance across a range of nystatin . . . . . . . . . . . . . . . . . . . . . . . . . . 603.3.5 Ergosterol phenotypes and map to fitness . . . . . . . . . . . . . . . . . . . . . . . 633.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643.4.1 Prevalence of sign epistasis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643.4.2 Maximum growth rate in one environment does not predict sterol phenotype orgrowth in other environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.4.3 Fitness landscapes in haploids and diploids . . . . . . . . . . . . . . . . . . . . . . 663.4.4 Implications for speciation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67viTable of Contents4 The limit to evolutionary rescue depends on ploidy in yeast exposed to nystatin . . . . . . . . 684.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704.2.1 Strains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704.2.2 Mutant acquisition in deep well boxes . . . . . . . . . . . . . . . . . . . . . . . . . 714.2.3 Confirming nystatin resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.2.4 Mutant acquisition with larger population sizes . . . . . . . . . . . . . . . . . . . . 734.2.5 Further testing of potential diploid mutants . . . . . . . . . . . . . . . . . . . . . . 734.2.6 Nystatin efficacy over time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 744.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754.3.1 Further testing of potential diploid mutants . . . . . . . . . . . . . . . . . . . . . . 764.3.2 Nystatin efficacy over time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 774.3.3 Rescue in larger populations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 774.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 825.1 Thesis summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 825.1.1 Chapter 2: Repeatability of adaptation . . . . . . . . . . . . . . . . . . . . . . . . 825.1.2 Chapter 3: Evolution of BDM incompatibilities between first-step adaptive muta-tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 835.1.3 Chapter 4: Limits to adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855.2 Putting adaptive mutations in the context of Fisher’s geometric model over environmentalgradients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 865.2.1 Can changing epistasis be explained by a shifting optimum? Revisiting Chapter 3 . 885.2.2 Can the fitness of a strain in an environmental gradient be predicted from its fitnessin a single environment? Revisiting Chapter 2 . . . . . . . . . . . . . . . . . . . . 915.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111A Appendix for Chapter 2: Too much of a good thing: The unique and repeated paths towardcopper adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111A.1 Supplementary Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111A.1.1 Quantitative real-time PCR (qPCR) . . . . . . . . . . . . . . . . . . . . . . . . . . 111A.1.2 Tetrad analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112A.2 Supporting Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114A.3 Supporting Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123viiTable of ContentsB Appendix for Chapter 3: Widespread Genetic Incompatibilities Between First-Step MutationsDuring Parallel Adaptation of Saccharomyces cerevisiae to a Common Environment . . . . . 128B.1 Supplementary Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128B.1.1 Strain construction details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128B.1.2 Segregating mutation in DSC2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130B.1.3 Preparing stocks for growth rate assays . . . . . . . . . . . . . . . . . . . . . . . . 130B.1.4 Analysis including outliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132B.2 Supporting Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133B.3 Supporting Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134C Appendix for Chapter 4: The limit to evolutionary rescue depends on ploidy in yeast exposedto nystatin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141C.1 Strain Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141C.2 Mutant Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142C.3 Supporting Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155C.4 Supporting Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156viiiList of Tables2.1 Mutations identified in the CBM lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.2 Oligonucleotides employed for real time PCR, Southern blot analysis, and genotyping . . . . 293.1 Beneficial mutations used for the study of epistasis . . . . . . . . . . . . . . . . . . . . . . 513.2 Results from a mixed-effects model using haploid maximum growth rate data in nystatin2 . . 583.3 Results from a mixed-effects model using homozygous diploid maximum growth rate datain nystatin2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.4 Results from a mixed-effects model using haploid maximum growth rate data in YPD . . . . 613.5 Results from a mixed-effects model using homozygous diploid maximum growth rate datain YPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61A.1 Date of isolation for CBM lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114A.2 T-test results comparing maximum growth rates between CBM lines and BY4741 in iron . . 116A.3 Predicted transcription factor binding site gains and losses from intergenic mutations . . . . 117A.4 T-test results comparing maximum growth rate between CBM lines and BY4741 in copper8 118A.5 T-test results comparing maximum growth rate between CBM lines and BY4741 in YPD . . 119A.6 T-test results comparing maximum growth rate between CBM lines and BY4741 in YPG . . 120A.7 Linear models of maximum growth rate of tetrads in YPD . . . . . . . . . . . . . . . . . . 121A.8 Linear models of maximum growth rate of tetrads in copper9, after correcting for growth inYPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121A.9 Additional mutations identified in small-colony forming CBM lines . . . . . . . . . . . . . 122B.1 DSC2 allele status in constructed haploid strains . . . . . . . . . . . . . . . . . . . . . . . . 130B.2 Experimental design of the growth rate assays . . . . . . . . . . . . . . . . . . . . . . . . . 133C.1 Results of ANOVA and post-hoc Tukey tests comparing mean number of days until growthbetween the different strain types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155ixList of Figures1.1 Histograms of absolute fitness deviations in chemicals investigated . . . . . . . . . . . . . . 111.2 Conserved genetic interactions between pairs of species . . . . . . . . . . . . . . . . . . . . 152.1 Observed mutations in copper-adaptation lines . . . . . . . . . . . . . . . . . . . . . . . . . 332.2 Copper tolerance across 34 copper-adapted lines . . . . . . . . . . . . . . . . . . . . . . . . 382.3 CBM growth rates under different environmental conditions . . . . . . . . . . . . . . . . . 392.4 Maximum growth rates of tetrad lines in copper9 . . . . . . . . . . . . . . . . . . . . . . . 402.5 Impact of mutations on growth of tetrad lines in copper9 . . . . . . . . . . . . . . . . . . . 413.1 Types of epistatic relationships between mutations . . . . . . . . . . . . . . . . . . . . . . . 473.2 Abbreviated ergosterol biosynthesis pathway . . . . . . . . . . . . . . . . . . . . . . . . . 503.3 Maximum growth rate of haploid strains in nystatin2 and YPD . . . . . . . . . . . . . . . . 573.4 Maximum growth rate of diploid strains in nystatin2 and YPD . . . . . . . . . . . . . . . . 593.5 Optical density after 24 hours of growth for haploid strains in a range of concentrations ofnystatin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623.6 Sterol profiles of allMATa haploid strains . . . . . . . . . . . . . . . . . . . . . . . . . . . 644.1 Visual representation of all mutant acquisition experiments . . . . . . . . . . . . . . . . . . 714.2 The number of potential mutants found over three acquisition experiments . . . . . . . . . . 764.3 Percentage of wells that grew within four days of inoculation in the nystatin efficacy experiment 785.1 Optimum overshooting in Fisher’s geometric model . . . . . . . . . . . . . . . . . . . . . . 875.2 Optima change with changing environments . . . . . . . . . . . . . . . . . . . . . . . . . . 885.3 Projecting mutational effects into one dimension . . . . . . . . . . . . . . . . . . . . . . . . 895.4 Changing epistasis in an environmental gradient . . . . . . . . . . . . . . . . . . . . . . . . 905.5 OD of copper beneficial mutation lines in a gradient of copper concentrations . . . . . . . . 92A.1 Optical density after 24 hours of growth of specific spores over a range of copper concentrations123A.2 Comparison of CUP1 copy number assays . . . . . . . . . . . . . . . . . . . . . . . . . . . 124A.3 Copper tolerance of knockout lines for identified genes . . . . . . . . . . . . . . . . . . . . 125A.4 Maximum growth rate of tetrads in YPD . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126A.5 Copper tolerance for specific spores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127B.1 Optical density for haploid strains in nystatin2 and YPD . . . . . . . . . . . . . . . . . . . 134xList of FiguresB.2 Optical density for diploid strains in nystatin2 and YPD . . . . . . . . . . . . . . . . . . . . 135B.3 Optical density for homozygous diploids in a range of concentrations of nystatin . . . . . . . 136B.4 Maximum growth rate of diploid strains for each gene combination in nystatin2 . . . . . . . 137B.5 Optical density for diploid strains in a range of concentrations of nystatin . . . . . . . . . . 138B.6 Maximum growth rate of haploid strains in nystatin2 and YPD when including outliers . . . 139B.7 Maximum growth rate of diploid strains in nystatin2 and YPD when including outliers . . . 140C.1 Growth curves of all populations from larger volume flasks . . . . . . . . . . . . . . . . . . 156C.2 Mean OD72 when populations are re-tested in nystatin versus the number of days untilgrowth was observed in the original acquisition experiments . . . . . . . . . . . . . . . . . 157xiAcknowledgementsI have to start by thanking my supervisor, Sally Otto. I was very lucky to be in your lab, and I can only hopethat others have the opportunity to have a supervisor like you. You have pushed me when I needed pushingand have been thoughtful and compassionate when I needed understanding. On top of that, I couldn’t haveasked for a better role model for how to be a scientist. Thank you for helping me to grow, both as an academicand as a person.I would also like to thank Aleeza Gerstein. I have been lucky to have you as a mentor throughout myPhD. You are a co-author on two of my chapters and a spiritual co-author on the third, and collaborating withyou has been one of the great joys of my PhD. Also, you give the best pep talks. Yeast friends 4 ever.Thank you to my committee, Dolph Schluter, Michael Doebeli and Corey Nislow. Your advice and inputhave helped to expand my thinking and improve my science.None of this would have been possible without all of the amazing people who’ve worked in the wet labover the course of my PhD. Thank you to Dara Lo, William Li, Angelica Lillico-Ouachour, Andy An, Li-WeiZhang, Ryan Stratychuk, and Chris James for their assistance in the lab, as well as Sherry Li and Anna VanTol for their work before I arrived. A special thanks to Marcus Campbell for helping me to figure out how tosupervise students and gain confidence in myself, Xinya Wang for bringing a steady and cheerful atmosphereto the lab, and to Lesley Miller for all of your help at the end of this long road. Of course, a big thank you toAnastasia Kuzmin, for all of your help but also for being there to talk to in the lab. Without you, I probablywould have gone crazy long ago.All of my lab mates have been great over the years, and I thank you all for your excitement and enthu-siasm, whether it’s for an epically long lab meeting or a ridiculous pumpkin carving. I especially want tothank Kate Ostevik, Nathaniel Sharp, Rich FitzJohn, Leithen M’Gonigle, Matt Osmond and Michael Scott.Being the only yeast lab in our building, the Vancouver Yeast Group has been an amazing resourcethroughout my PhD, and many experiments would have been much more difficult without their help. Aspecial thank you goes out to Krystina Ho, Vivien Measday and her lab, Chris Walkey and Allan DeBonofor sharing their expertise and resources. Patricia Schulte and Tim Healy also provided expert assistance andadvice when sorely needed.A big thank you to everyone both in the Zoology Department and in the Biodiversity building who makegrad school such a great place to be. Everyone who’s ever organized a discussion group, a seminar speaker,a party or a symposium, or taken part in a skit, you are amazing. I especially want to thank some of thepeople who’ve made my own experience at UBC a special one including Kieran Samuk, Greg Owens, AlanaSchick, Milica Mandic, Megan Bontrager, Kristin Nurkowski, Gina Conte, Brook Moyers, Kathryn Turner,Matt Siegle, Matt Pennell, Sam Yeaman, J.S. Moore, Anne Dalziel, Julie Lee-Yaw, and Kim Gilbert.I also thank the funding sources that made these projects possible: the Four-Year fellowship from UBCxiiAcknowledgementsand funding from the National Science and Engineering Research Council of Canada (Discovery Grant toS.P. Otto).Finally, a thank you to my family, to whom this thesis is dedicated, as well as to all of my ‘unofficial’family: Courtenay Brown, Brett Willis, Vicki Lee, Nathaniel Brown, Kieran Bol, Grady Johnson and RyanVanderhoek. Thank you for still liking me even though I never email/text/call back. And, of course, to DanSchick. You’ve been there for me through the whole thing. You’ve accepted my weird hours, messy officeand disdain for dishes. It’s hard to explain how much I appreciate you.xiiiDedicationThis thesis is dedicated to my parents, George and Lulu, for their continued support, even if they didn’t knowwhat I was doing or why I was doing it.xivChapter 1IntroductionWe have come a long way since the time of Charles Darwin and Alfred Russel Wallace in our understandingof evolution (DARWIN 1859; DARWIN and WALLACE 1858). During the Modern Synthesis (roughly 1930s-1940s), ideas were developed about the interplay between genetics and evolution, and we have been testingand refining those ideas in ways that were unimaginable to those original scientists. Investigations of biolog-ical systems, both in the wild and in the lab, provide an expanding collection of case studies, which informus about what has happened and how it has happened in specific instances. Evolutionary theory helps us todetermine what patterns might arise and explores what might be possible, given a certain set of conditions.Experimental evolution borrows elements of both and can be found somewhere in between. In experimentalevolution, we collect case studies of what actually happens given a certain set of conditions. Having control(although imperfect) over these conditions allows us to understand how slight differences in experimentalinput can lead to differences in biological output. Importantly, experimental evolution can tell us about therepeatability of such outcomes when the experiments are performed in replicate. Repeatability speaks to theinner workings of the evolutionary process, revealing how specific conditions interact with available geneticvariation (whether from standing genetic variation or mutation) to influence the path taken. Data from otherfields, especially molecular biology, can aid in the interpretation of the observed evolutionary paths, allow-ing us to develop mechanistic models to explain why certain outcomes should occur. Finally, the increasingabundance of biological data from all fields, and its organization into well-maintained databases, has mademeta analysis an increasingly useful tool for developing, exploring and testing evolutionary hypotheses.Throughout this thesis, I use experimental evolution with the model yeast Saccharomyces cerevisiae tostudy the processes of adaptation and speciation. My primary focus is on the underlying genetic changes,and that will be the focus of this chapter. Genetic systems are complex, and understanding how a genotypeis translated into a phenotype is a central problem in biology. For this reason, the constraints and limitationsthat shape evolution are very difficult to predict or model a priori, and the genetic system is often the sourceof many of the surprises when conducting experiments.The benefits and limitations of experimental evolution have been explored in previous reviews (e.g.,BAILEY and BATAILLON 2016; BARRICK and LENSKI 2013; KAWECKI et al. 2012; LONG et al. 2015) andbooks (GARLAND and ROSE 2009), with some particularly focussed on microbes (ADAMS and ROSEN-ZWEIG 2014; KASSEN 2014; LENSKI 2017). One major advantage of using a model organism such as S.cerevisiae for our experiments is that we have the ability to delve deeply into the underlying genetic basisof the observed outcomes, asking and answering genetic questions about evolution that would otherwise bevery difficult to address. In this thesis, I hope to contribute to our collective understanding of evolution,especially as it relates to the genes involved in adaptation and how they might act given their specific genetic11.1. Genetic model organisms and the study of evolutionand environmental context. I will explore how different factors can constrain and limit genetic adaptationand potentially contribute to speciation including: adaptive environment (both the qualitative nature of theenvironment and the exact concentration of a stressor), epistasis between adaptive mutations, ploidy levelof an organism, and the interplay between these factors. These insights will help to inform future studieson organisms with more difficult to decipher genetic systems, including those from natural populations, byproviding candidate explanations for observed phenomena that can be specifically targeted for testing.In this Introduction chapter, I will first discuss genetic model organisms and their usefulness for the studyof evolution as a backdrop for the rest of the thesis. I will then discuss the repeatability of adaptation andhow genomic breadth (how many potential paths evolution might take) and mutation types (and their relativerates) might affect repeatability. Genetic interactions between potentially adaptive mutations can furthershape the repeatability of adaptation if early adaptive mutations change the fitness effects of later mutations,directing evolution down certain paths that depend on those early stochastic events. These interactionsresult in lower repeatability than if all mutations had fixed effects, and also have implications for speciationamong populations diverging in separate geographic areas (allopatry). Finally, I will discuss the limits ofadaptation, as determined by the genomic breadth of adaptive mutations in a specific environment and bythe availability of such mutations to the organism in question, before moving on to explore how large-scaledatasets from molecular biology can potentially help us to interpret evolutionary data and hypothesize aboutpossible evolutionary trajectories.1.1 Genetic model organisms and the study of evolutionModel organisms are useful in the study of adaptation and speciation, especially when we are specificallyinterested in uncovering the genetic and molecular basis. By using the same model organisms for studiesof evolution as for molecular biology, we can take advantage of the information and perspectives gatheredfrom years of study by hundreds of scientists. For example, many genes have been characterized in thesemodel organisms, and information about them has been organized into openly-available databases (yeast:Saccharomyces Genome Database, or SGD, CHERRY et al. 2011; Caenorhabditis elegans: WormBase,STEIN et al. 2001; Drosophila melanogaster: FlyBase, GRAMATES et al. 2017; Escherichia coli: EcoCyc,KESELER et al. 2017 and others; Arabidopsis thaliana: The Arabidopsis Information Resource, or TAIR,HUALA et al. 2001). In addition, these model organisms tend to have small, easily sequenced genomes withhigh-quality reference genomes, so candidate causative mutations can be found with relative ease. Becauseof the availability of molecular information, it is easier to filter this list of mutations by gene function to findthose that are most likely to be responsible for the phenotype of interest. Finally, the many genetic tools andtricks that have been developed for these organisms simplify testing of specific genetic hypotheses.While each organism can be used for a variety of questions, some are better suited for certain interests.For example, Caenorhabditis species are a good choice for studies about the evolution of sex, because onecan manipulate the amount of outbreeding present (CUTTER 2005). D. melanogaster is well-suited to studythe dynamics of evolution with obligate outcrossing (BURKE et al. 2010). And if an aspect of plant evolutionis of interest, A. thaliana is the natural choice (MAURICIO 1998). Yeast is well-suited for studies where we21.2. Repeatability of evolutionwish to characterize the nature of the genes involved in evolution, including but not limited to their identity.S. cerevisiae is the best genetically characterized eukaryote and has easily manipulated genetics (BOTSTEINand FINK 2011). In yeast, it is standard practice to insert specific mutations into known locations of thegenome by transformation and perform controlled crosses where individual meiotic progeny can be dissectedout and grown separately (SHERMAN 2002). It was for these reasons, and because of their fast generationtime for a eukaryote, that I chose to study yeast.Yeast is an increasingly popular model system for experimental evolution and evolutionary genetics,including speciation research. Their rapid generation time (replicating in ~90 minutes under ideal conditions,SHERMAN 2002) allows for evolution experiments to be conducted efficiently. Yeast primarily reproduceasexually, either in a haploid or a diploid state, and can do so in either liquid or solid medium. While manystrains of S. cerevisiae form clumps in liquid medium (e.g., brewing strains, SOARES 2011), lab strains havebeen selected to have dispersed cells (MORTIMER and JOHNSTON 1986), facilitating the quantification ofabundance through the use of optical density measurements. Fast, asexual growth on agar allows for easyphenotyping and for bottlenecking of populations down to individual genotypes, the progeny of which canbe used in fitness assays or for DNA extraction. Yeast can also be preserved in a frozen state, allowingfor the long-term storage of strains, and their well-annotated, small genome (12 Mb) is easily sequenced(SHERMAN 2002). Although the ‘natural’ ecology of yeast is largely unknown, the many advantages ofyeast as a genetic system have prompted studies into the evolution and ecology of natural yeast (as reviewedin LANDRY et al. 2006, REPLANSKY et al. 2008, LITI 2015). Studies of natural yeast complement labinvestigations, developing yeast into a more complete model for studies of evolution.1.2 Repeatability of evolutionStephen Jay Gould proposed a thought experiment in his book Wonderful Life in which we might learnabout the repeatability, and therefore predictability, of evolution by “replaying life’s tape” and looking at thesimilarity of the outcomes (GOULD 1989). For natural populations, the closest we can come to replayingevolution is observing cases of parallel adaptation, where adaptation has “played” multiple times in separatebut similar environments. From these populations, we know of many examples of phenotypic repeatabilityincluding the repeated loss of lateral plates when stickleback move from the marine environment into fresh-water (SCHLUTER et al. 2004), the evolution of species of Anolis lizards with similar niches, morphologyand behaviour (ecomorphs) on separate islands (LOSOS 1992) and the development of a similar gene expres-sion pattern among clinical isolates of Pseudomonas aeruginosa from patients with cystic fibrosis (HUSEet al. 2010). Some cases of parallel phenotypic evolution are also underlain by parallel genotypic evolution,like wing pigmentation patterns in male Drosophila that involve regulatory changes of the same gene (yel-low) (PRUD’HOMME et al. 2006) and stickleback morphological trait divergence in species pairs (CONTEet al. 2015), but others are not, like beach mice adapting to sandy coastal dunes in Florida’s Gulf and Atlanticcoasts (MANCEAU et al. 2010). I will focus the rest of this discussion on genetic repeatability, as that is themain focus of Chapter 2.The generality of genetic repeatability in natural systems can be studied using meta-analyses, such as31.2. Repeatability of evolutionthat performed by CONTE et al. (2012). In this study, the authors estimated the probability of gene reusefor natural parallel phenotypic evolution and found that the mean probabilities were relatively high (0.32 forstudies where the whole genome was considered when mapping phenotypes, 0.55 for studies where only can-didate genes were tested for association with the phenotype). They concluded that the biases and constraintsof evolution were quite strong, restricting the paths available to evolution and resulting in high repeated geneuse. Gene reuse was also higher for more closely related populations, which can be interpreted as either asimilarity in their genomic biases, in the standing genetic variation available to them, or a combination ofthe two. Truly parallel environmental conditions are difficult to verify in nature, however, where there areseemingly endless sources of ecological variation experienced by populations that could all affect the proba-bility of certain evolutionary outcomes. If a pair of natural populations adapt differently to seemingly similarconditions, we cannot be sure that there was not a key unmeasured difference between their environmentsthat caused their divergence, and this problem only increases with increasing numbers of populations (and,therefore, environments) included.Experimental evolution allows us to “play life’s tape” multiple times for a single set of conditions, con-currently, and compare the results. In such studies, different degrees of repeatability have been observed. Ina study by MEYER et al. (2012), phage l, which infect E. coli, were evolved and many developed the abilityto utilize a new receptor, OmpF. All adaptive mutations were in the virus’ J protein, and the phage that wereable to use OmpF all had four mutations in common: two identical mutations across strains, a mutation in oneof two adjacent positions and a mutation within a certain 30-basepair block. Another study by LOURENÇOet al. (2016) found seven genes targeted in parallel instances of adaptation of E. coli during gut colonizationof mice. In yeast, a study by LANG et al. (2013) found a small subset of genes that drove increases in fitnessin rich medium that were repeated among replicates (observed in 3 - 21 replicate populations out of 40).Similarly, in yeast adapting to a rich medium, KRYAZHIMSKIY et al. (2014) observed some genes to evolverepeatedly to varying degrees (each in 3 - 12 out of 104 total populations), and the proportion of gene usevaried among starting genetic backgrounds. Many more examples of repeated evolution from experimentalevolution studies have been reviewed in LOBKOVSKY and KOONIN (2012).From these studies, we know that repeatability varies, but it is still unclear how it will vary depending onthe environmental conditions. We know that it should vary with the genomic breadth of potentially beneficialmutations, which is itself tied to the environment in which adaptation occurs (e.g., GRESHAM et al. 2008).If many mutations are potentially beneficial in an environment, then there are many potential paths thatevolution can follow. If, in contrast, only a few specific mutations will increase fitness of an organismin an environment, then adaptation will be highly repeatable due to this constraint. Repeatability shouldalso vary with the types of mutations that are potentially beneficial and their relative mutation rates. Forexample, if both additional chromosomal aneuploidy and specific SNPs in a single gene increase fitness in anenvironment but aneuploidy occurs more frequently then we expect evolution to use aneuploidy more oftenbecause of the increased opportunity. In Chapter 2, I explore the effects of environment and mutation type ongenetic repeatability in parallel populations adapted to high concentrations of copper. By testing individualfactors (such as environment type) for their role in repeatability as well as considering genetic results withinthe context of broader biological knowledge (mutation types, molecular or biochemical characterization), we41.3. Genetic interactionscan begin to understand the mechanics of repeated evolution.1.3 Genetic interactionsGenetic interactions, measured as epistasis between alleles, also shape the outcomes, and therefore repeata-bility, of evolution. The word “epistasis” has a few different meanings in genetics and evolution, althoughthese meanings are related (ROTH et al. 2009). Originally coined by BATESON (1909), epistasis was used todescribe the interaction between genes in which the action of one gene was blocked by the action of another.This is how molecular geneticists still use the word, and they use epistasis analysis to determine the orderof action of genes in regulatory pathways (AVERY and WASSERMAN 1992). Evolutionary geneticists, onthe other hand, use the word to broadly describe any form of gene (or allele) interaction deviating from theexpectation of independent effects (WHITLOCK et al. 1995). Depending on the appropriate scale for thephenotype being measured, either additive or multiplicative interactions can be the expectation.In evolutionary biology, the ultimate phenotype of interest is usually fitness. Fitness, and traits often usedas a proxy for fitness such as growth rate, are composite traits, underlain by many other biological processes.It is easy for epistasis to arise as a consequence of the many potential interactions between these underlyingbiological traits in determining the ultimate fitness of the organism. For example, if fitness depends on theproducts of a metabolic pathway and flux through that pathway is optimized at an intermediate value, thefitness effects of mutations in the enzymes of the pathway will depend on the flux rate of the whole pathwayrelative to the optimum (SZATHMÁRY 1993). If flux were below the optimum, a small effect mutation thatincreases flux would be beneficial. If there were a larger effect mutation that increases flux slightly above theoptimum, this mutation may also be beneficial and spread in the population. But, the second mutation wouldmake the first flux-increasing mutation deleterious in the new genetic background. This example is similarto any other trait undergoing stabilizing selection and will result in epistasis for fitness among loci affectingthe trait, even if those loci act additively on the underlying trait (WHITLOCK et al. 1995).The nature of epistasis between potentially beneficial alleles can determine the speed of adaptation orconstrain evolutionary trajectories. Positive epistasis between beneficial alleles, where the two mutations to-gether are fitter than the expectation, can theoretically increase the rate of adaptation, but few cases of positiveepistasis have been found for beneficial alleles (BUSKIRK et al. 2017; CHOU et al. 2009). Negative epistasisis more commonly observed, where new beneficial alleles have smaller effects when in the same geneticbackground as other beneficial alleles, slowing the rate of adaptation (CHOU et al. 2011; KRYAZHIMSKIYet al. 2014). When epistasis becomes so negative that adding the second mutation is actually detrimental in-stead of beneficial (“sign epistasis”), the evolutionary trajectory of a population can be strongly constrained(WEINREICH et al. 2005). An example of this was found by TENAILLON et al. (2012) who evolved 115populations of E. coli to high temperature. They found repeated evolution at the gene level, but there wereat least two distinct pathways available, involving either the RNA polymerase complex or the terminationfactor rho. Early adaptive mutations were inferred to have directed later adaptation of the populations intoone of these two paths through negative epistasis between these early mutations and subsequent ones. In thisway, epistasis affects the amount of repeatability observed between populations since repeatability will be51.3. Genetic interactionslower when there are mutually-exclusive solutions for evolution to explore than when all beneficial mutationsmaintain their fitness effects regardless of genetic background. In the case of full independence, we wouldexpect all populations to eventually converge upon the single best genotype, even with initial stochasticdivergence.1.3.1 Bateson-Dobzhansky-Muller model of speciationThe effects of epistasis not only shape adaptation within a population, but they also influence the outcomes ofhybridization between populations in the main genetic model of speciation, the Bateson-Dobzhansky-Muller(BDM) model. In the BDM model of speciation, alleles that are beneficial or benign in their normal geneticbackground cause sterility or inviability in hybrid individuals (ORR 1995). Such interactions are known asBDM incompatibilities. The alleles are able to spread in different sub-populations because they do not causereductions in fitness in those sub-populations’ genetic backgrounds. Then, when the sub-populations arebrought together and hybrids are formed, the alleles are tested together in novel genetic combinations andhybrid individuals have lower fitness than either of the parental types, reflecting negative sign epistasis. Lowhybrid fitness results in decreased gene flow between sub-populations and, therefore, reproductive isolation.Despite interest in BDM incompatibilities, only a few examples have been characterized at the molecularlevel (PRESGRAVES 2010). Most of these are found between species adapted to different local environmentsand the causative alleles are presumably beneficial in those separate environments (documented in NOSIL andSCHLUTER 2011). Other genetic incompatibilities depend on the environment in which they are measured,as have been found in natural populations of yeast (including one characterized two-locus BDM, HOU et al.2015), but it is unknown whether the evolution of these alleles is linked to the external environment sincewe have no concrete knowledge of the evolutionary history. Additionally, some cases of incompatibilitybetween natural populations have no clear connection to the external selective environment (see examples inMAHESHWARI and BARBASH 2011).Experimental evolution studies allow direct control over the environment, and incompatible mutationshave been found in some cases. DETTMAN et al. (2008) found reduced reproductive success in matingsof lineages adapted to different environments compared to matings between lineages adapted to the sameenvironment in Neurospora crassa, consistent with the action of BDM incompatibilities. They found similarresults when mating populations of S. cerevisiae that had adapted to different environments (DETTMAN et al.2007). In the latter case, the underlying BDM incompatibility was subsequently mapped (ANDERSON et al.2010) and was the first reported BDM interaction among known genes that was isolated from experimentallyevolved strains, to my knowledge. An incompatibility has also been found among experimentally evolvedpopulations adapting to the same environment in S. cerevisiae (KVITEK and SHERLOCK 2011). There isanother case of an experimentally evolved incompatibility arising from parallel selection in Methylobac-terium extorquens (CHOU et al. 2014), but the nature of species (and reproductive isolation) is less clear fororganisms without meiosis thus making it difficult to classify this interaction as a BDM incompatibility.The relatively rapid (within 448 mitotic generations in the case of KVITEK and SHERLOCK 2011) es-tablishment of alleles leading to BDM incompatibilities in the above experiments can be attributed to theaccumulation of alleles driven to high frequency by selection. There are two main mechanisms of speciation61.4. Limits to adaptationby selection, referred to as ecological and mutation-order speciation (SCHLUTER 2009). Ecological specia-tion occurs when the divergence between groups is driven by divergent selection in different environments(SCHLUTER 2009). Since selection is divergent, it drives the fixation of different alleles in each group, eachadvantageous in the local environment but not necessarily in the other. This process can lead to the evolu-tion of any type of reproductive isolation, including premating isolation, hybrid sterility and intrinsic hybridinviability as well as extrinsic, ecologically based pre- and postzygotic isolation (SCHLUTER 2009). In con-trast, mutation-order speciation involves divergence between groups that occurs as a by-product of differentmutations arising and fixing in separate groups adapting to similar selection pressures (SCHLUTER 2009).The evolution of reproductive isolation occurs by the chance fixation of different advantageous mutationsin different groups, even though the same mutations would be initially favoured in both (SCHLUTER 2009).Early divergent adaptive mutations lead to increased divergence between groups due to the effects of epistasisinfluencing the evolutionary trajectories. It is important to emphasize that, even in identical environments,selection can become divergent between groups when the genetic background changes, thereby changing themutations that confer a fitness benefit. Intrinsic reproductive isolation can then arise as a consequence ofincompatible genetic solutions.The aforementioned studies demonstrate that BDM-type interactions can establish over the course ofhundreds of generations in experimental evolution studies by either mechanism of speciation by selection.What remains unknown thus far from long-term experiments of populations evolved under the same selectivepressure is how frequently first-step adaptive mutations themselves could contribute to reproductive isolation.It is possible that incompatibility starts at the first steps of divergence in populations adapting in parallel, butit is also possible that those early stage mutations only change the selective environment of future mutations.I investigate epistasis between first-step adaptive mutations in Chapter 3 using fungicide-adapted mutations.I find that BDM incompatibilities are fairly common between these different, large effect adaptive mutations,demonstrating that mutation-order speciation may arise even at the very first step of adaptation.1.4 Limits to adaptationIn extreme cases of evolutionary constraint, populations will not be able to respond genetically to the en-vironment and will have reached their adaptive limit. Such adaptive limits are thought to contribute to thelimited geographic distribution of species, and have been suggested by observed trade-offs between traitsimportant for adaptation to different edges of a species’ range (for example, ANGERT et al. 2008). The lim-its to adaptation include not only whether a beneficial mutation is possible, but also whether it will be ableto successfully rise in frequency. This depends on a host of population genetic factors, most of which I willnot address here, but have been well reviewed in BARTON and PARTRIDGE (2000). In this section, I willprimarily focus on the availability of genetic variation for selection to act upon. Lack of genetic variationin the trait of interest is a common explanation for the observation of a limit to evolutionary change (BELL2013; BLOWS and HOFFMANN 2005). Most papers on the topic are primarily concerned with natural pop-ulations, in which standing genetic variation is the primary source of variability. In experimental evolutionstudies, unless standing genetic variation is of explicit interest, the populations are typically started with as71.5. What can large-scale datasets tell us about evolution?little variation as possible, coming from either inbred lines or originating from single clones. In these exper-iments, genetic variation still determines the rate of initial response to selection (BARTON and PARTRIDGE2000), but it is limited by the input of spontaneous mutation (DE VISSER and ROZEN 2005).When considering a specific trait under selection, or a specific environment to which a population istrying to adapt, the beneficial input of mutation can be limited in a few different ways. One well-discussedlimit is imposed by mechanistic, physiological or developmental constraints of the organism, where certaintypes of variation are not possible (BLOWS and HOFFMANN 2005; SMITH et al. 1985). The same epistaticrelationships that constrain evolution to follow certain trajectories dependent upon initial beneficial mutationscan also limit further evolution entirely in extreme cases (SMITH et al. 1985). This arises as a result ofsign epistasis between the current genetic background and all potential beneficial mutations, making thesemutations inaccessible. These factors affect the mutational target size for adaptation. In a given environment,if the genomic breadth of potentially beneficial mutations is non-existent, the organism in question will not beable to adapt and will have met its adaptive limits. Another possibility is that a population is instead limitedby the mutation rate (BLOWS and HOFFMANN 2005). If potentially beneficial mutations are possible but areexceedingly rare due to very low mutation rates for those specific genomic changes, it is very unlikely that apopulation will be able to adapt. It will have effectively met its adaptive limits and, if it has a negative growthrate, will go extinct before being rescued by evolution. Because the genomic breadth of adaptation and therate of beneficial mutation are decided by the environment in question and the biology of the organism,changing either can cause an organism to reach its adaptive limit. In Chapter 4, I look at ploidy as an intrinsicbiological feature that can affect the accessibility of beneficial mutations. Ploidy is not often consideredamong the factors thought to influence the limits of adaptation, but both the effects of mutations (related togenomic breadth) and the rates of certain mutations (especially certain mutation types) can differ by ploidy,even in otherwise identical genomic backgrounds. It is thus important to determine how ploidy can affectevolutionary dynamics, especially when considering the evolution of antibiotic resistance, for which bothhaploids and diploids are common targets.1.5 What can large-scale datasets tell us about evolution?Forward genetic screens have been used to determine the genetic basis of phenotypes since the 1940s (BEA-DLE and TATUM 1941). In these studies, collections of mutants, usually generated by a mutagen, are screenedfor a phenotype of interest thought to be related to a particular biological process. All mutants displaying thephenotype of interest are characterized, both genetically and phenotypically. Genetic characterization caninvolve sorting the mutants into complementation groups (where each group is likely to represent a singlegene), genetic mapping of causative mutations, and sequencing of those mutations. Phenotypic character-ization depends on the biological process of interest, but could involve morphological measurements, de-termination of cellular localization patterns, and biochemical measurements of underlying reactions. Muchof our knowledge about the functions of individual genes was collected in this way. Then, to order thesegenes into genetic pathways leading to the phenotype of interest, double mutants were created and epistaticanalysis was performed. In molecular genetics, this type of work has been complemented by the use of81.5. What can large-scale datasets tell us about evolution?mutation collections and large-scale screens, which has greatly expedited the collection of biological data.In this section, I would like to explore what we, as evolutionary biologists, might learn from these large-scale datasets. This data has been primarily applied to models predicting the evolution of microbial systemsand has been reviewed elsewhere (PAPP et al. 2011), but I would like to broaden the view to include othernatural systems. Specifically, I am interested in the genetic interaction datasets, which tell us about epistasisbetween large collections of mutations. I see there being three major barriers to using this kind of data tounderstand evolution. First, these experiments are often performed with knockout mutations, but does thiskind of mutation reflect common natural mutations? Second, how generalizable are the specific interactionresults across different levels of biology (different populations, genetic backgrounds, species, genera, etc.)?And third, how do these genetic interactions depend on the environment in which they are measured? In thefollowing sections, I consider different data sets that provide insight into the answers to these questions.1.5.1 How well do mutation collections represent natural mutations?Each of the model genetic organisms explored above has a large-scale mutation collection (yeast: deletioncollection, GIAEVER et al. 2002; C. elegans: deletion collection, C. ELEGANS DELETION MUTANT CON-SORTIUM AND OTHERS 2012, and million mutation project, THOMPSON et al. 2013; D. melanogaster: genedisruption project, BELLEN et al. 2011; E. coli: Keio deletion collection, BABA et al. 2006; A. thaliana:T-DNA insertion collection, ALONSO et al. 2003). The collections differ in scope (proportion of the genescovered) and mutation type, but perhaps the best collection is in yeast. The large-scale whole-gene deletioncollection in S. cerevisiae includes all viable gene deletions and all essential genes deleted in a heterozygousstate (GIAEVER et al. 2002), as well as temperature-sensitive alleles for the essential genes (COSTANZOet al. 2016). In addition, there are collections of plasmids available in yeast for overexpression studies (highcopy plasmids) (JONES et al. 2008) and complementation mapping of genes (low copy plasmids) (HO et al.2009; HVORECNY and PRELICH 2010). Together, these collections have been used to characterize generalproperties of the cell (e.g., finding the number of essential genes) as well as the functions of individualproteins and their interactions, but how well do these mutations represent naturally occurring mutations?A study by DOWELL et al. (2010) found that, even for whole-gene deletions, the effect of some mutationscan depend on their genetic background. When they systematically deleted genes in a second strain of S.cerevisiae, 57 genes were only essential in one of the two backgrounds (DOWELL et al. 2010), indicatingthat results from large-scale deletion collections may not be universally applicable, especially for less severephenotypes. However, a study by PAYEN et al. (2016) used single-gene deletions and amplifications toidentify potentially beneficial mutations in three environments (limitation of phosphate, glucose or sulfate)by their fitness effects in pooled competitions. When comparing the results to sets of mutations found duringexperimental evolution in the same conditions, they found that, on average, ~35% of the mutations thatoccurred during experimental evolution were predicted to be beneficial by the systematic screen (PAYENet al. 2016). In addition, a study by JELIER et al. (2011) found that they were generally able to predict thephenotypes of a variety of yeast strains based on their genome sequences. They first predicted the impactof individual mutations within genes on protein function, focussing on loss-of-function mutations. Then,utilizing high-throughput data sets containing the growth rates of the deletion collection under different91.5. What can large-scale datasets tell us about evolution?environmental conditions acquired from the SGD (CHERRY et al. 2011), they estimated the relevance ofeach protein perturbation to growth in each condition. From this, they predicted the growth of each strain ineach condition based on the strain’s mutated genes, and they assessed the performance of their predictionsby measuring growth across 20 conditions.The ability to predict the potential fitness effects of loss-of-function type mutations en masse is amazingin and of itself, but can we do the same kind of analysis without limiting it to predicted loss-of-functionmutations? I have attempted to begin to investigate this question by taking sets of genes found to impactresistance to chemicals from EHRENREICH et al. (2012) and determining whether these genes also impactfitness, on average, when deleted as either a heterozygote or homozygote, in a study by LEE et al. (2014).EHRENREICH et al. (2012) used a technique that they termed “extreme QTLmapping” to identify the geneticbasis of resistance to 13 chemicals in segregants of all pairwise crosses of four ecologically and geneticallydiverse yeast strains. In doing so, they detected more than 800 loci with an effect in at least one of the drugs.From their set of 13 chemicals, I identified five chemicals that were also tested in the screens of LEE et al.(2014), one of which had been tested twice at two different concentrations. LEE et al. (2014) tested ~1,100yeast strains heterozygous for a deletion in an essential gene and ~4,800 yeast strains homozygous for adeletion in a nonessential gene for their response to 3,250 chemicals that inhibit wild-type growth.If both natural mutations and gene deletions uncover roles for genes in certain environmental conditions,then we would expect the same genes to be implicated in fitness deviations in both studies. I will use thecandidate genes from EHRENREICH et al. (2012) to determine which genes out of the deletion set of LEEet al. (2014) are predicted to have large fitness deviations in certain chemicals. For all chemicals that weretested in both experiments, I retrieved the list of candidate genes that putatively affect resistance to thatchemical from EHRENREICH et al. (2012). I then downloaded all of the fitness data from LEE et al. (2014)for those chemicals. Because fitness data was reported as either negative or positive (having a growth defector growth advantage in the chemical when compared to the control), I took the absolute value of all fitnessdeviations to test only whether the genes had an effect, ignoring the direction of that effect. This shouldinclude cases where the deletion of a gene has a negative effect on fitness but an overexpression allele of thesame gene (which is possibly present among the natural variants) has a positive effect. In order to determinewhether candidate genes had, on average, a larger effect on fitness than expected in a certain chemical, Itook 10,000 random samples (without replacement) of the same size as the number of candidate genes fromthe full dataset for that chemical. I then compared the mean fitness deviation of the candidate genes to thedistribution of means calculated from the samples. I found that the candidate genes only had a larger averageeffect than expected by chance (>95% of samples) in two cases: in 4-nitroquinoline 1-oxide (4-NQO) and inone of the two experiments run with tunicamycin (Fig. 1.1). Of the two concentrations of tunicamycin testedin LEE et al. (2014) (25 nM and 200.47 nM), the candidate genes had a larger fitness effect compared to theaverage in the higher one (200.47 nM), which is closer to the concentrations used by EHRENREICH et al.(2012) for their selective plates (between 2.5 mM and 3.5 mM).From this analysis, we conclude that genes responsible for natural variation in an environment do notcorrespond well with those whose deletions show the largest fitness response in that environment. Thislack of correspondence can be partially attributed to the differences in methods between the two studies101.5. What can large-scale datasets tell us about evolution?4−NQO0 1 2 3 4 50100300500 p = 0.04*Benomyl0 1 2 3 4 50100300500 p = 0.44Chlorpromazine0 1 2 3 4 50100300500 p = 0.52Cycloheximide0 1 2 3 4 50100300500 p = 0.16Tunicamycin (25 nM)0 1 2 3 4 50100300500 p = 0.31Tunicamycin (200.47 nM)0 1 2 3 4 50100300500 p = 0.04*FrequencyAbsolute fitness deviationFigure 1.1: Histograms of the distributions of absolute fitness deviations in each chemical investigated (datafrom LEE et al. 2014). X-axes are cut off to exclude extreme values but include at least 99% of the data.Above each histogram, the means of the whole dataset (in grey) and the candidate genes for that chemicalfrom EHRENREICH et al. (2012) (in black) are shown. P-values are calculated as the proportion of randomlysampled means (out of 10,000 samples) that are greater than the mean of the candidate genes in that chemical(one-tailed test). Significant p-values are in bold with an asterisk (*).111.5. What can large-scale datasets tell us about evolution?being compared. Fitness was assayed differently in the two experiments, where EHRENREICH et al. (2012)used plates containing chemicals to select for highly resistant segregants and LEE et al. (2014) extractedfitness data from yeast growing in liquid medium. LEE et al. (2014) grew pooled samples of yeast in liquidmedium that contained the chemical of interest, where each pool consisted of either all heterozygous or allhomozygous strains. Because each deletion is uniquely barcoded, the fitness defect of each strain in thepool was measured by the relative abundance of its barcode in sequenced pools from the treatment samplescompared to control samples (without chemicals), allowing for a quantitative measurement of fitness. Thedifferent types of medium used between studies, in addition to differing concentrations of chemical usedboth within the study done by EHRENREICH et al. (2012) (different selective concentrations were used fordifferent crosses) and between the two studies, could potentially select for mutations in different genes. Thiswill be a persistent problem in using molecular biological data to interpret evolution, however, because wecannot expect methods to be the same unless the studies are conducted for the purpose of comparison.In addition, due to the nature of QTL mapping, EHRENREICH et al. (2012) could not identify thecausative gene in most cases, but instead mapped the phenotypes to small windows containing up to 14candidate genes, after excluding genes that contained no segregating polymorphisms. This means that thegenes included in our analysis may not be causative in the original crosses of EHRENREICH et al. (2012).Again, this will be a persistent problem with QTL analyses, where we might have to restrict ourselves tousing large-scale data to help narrow down candidate gene sets to those most likely to be involved. Thecandidate genes from EHRENREICH et al. (2012) were also mapped based on naturally-occurring variation,as opposed to the deletion mutations used in LEE et al. (2014). This was the key comparison that I wasinterested in making because evolutionary data will consist of these naturally-occurring mutations, but weexpect some types of natural mutations (like gain-of-function alleles) to be especially poorly represented bydeletion mutations. The mutations from EHRENREICH et al. (2012) were also mapped in haploid popula-tions, as opposed to the diploids used in LEE et al. (2014). If mutations only have an affect on the phenotypein one of the two ploidies, they will only be observed in one of the two screens. Additionally, for all essentialgenes (those that cannot be fully knocked out), only heterozygous mutants were tested in LEE et al. (2014),while haploid mutations in these genes could exist among the set screened by EHRENREICH et al. (2012),if non-knockout mutations are tolerated. I also included candidate genes found in any genetic backgroundby EHRENREICH et al. (2012), not limiting the set to those found in the BY background, which is sharedby the deletion mutants of LEE et al. (2014). Based on the poor predictive power observed in our compari-son within a single species, it seems unlikely that we can currently use single large-scale datasets to predictphenotype-causing alleles in other, more phylogenetically distant, organisms.Regardless, it is encouraging that I found any correspondence at all in this small comparison, and theresults of JELIER et al. (2011) give me confidence that collecting more of these large-scale datasets andperforming more of these kinds of comparisons will lead to a better ability to infer phenotypic causalityfrom genomic data of phenotyped individuals in the future. Although, in this analysis, I have used mapped,natural variants to predict the phenotypes of deletion mutants, I imagine that this kind of comparison couldbe done in the opposite direction to identify candidate natural variants involved in producing a phenotype ofinterest. If, for example, we were trying to determine the alleles involved in copper tolerance of an organism121.5. What can large-scale datasets tell us about evolution?and we had the relevant genomic data, we might look at a dataset of deletion mutant phenotypes in copper,like those produced by LEE et al. (2014), and identify the deletions with the highest fitness deviations. If anyof the corresponding genes are mutated in our organism’s genomic data (especially when compared with aclose relative that differs in its copper tolerance), then those genes might be implicated in copper tolerancefor this organism. This method could be used as a supplement to QTL mapping or to inform candidate geneanalysis. We only based our results on one set of data, unlike JELIER et al. (2011) who had multiple datasetsavailable to them per environment in some cases. For these environments, they were able to choose the morereliable dataset, based on the connectivity of the implicated genes in a predicted functional gene network foryeast (YeastNet, LEE et al. 2007). By continuing to collect high-throughput datasets, we can improve on thequality of information available for the phenotypic effects of genic mutations. It would be especially usefulif future high-throughput studies collected data from other types of mutations (such as using overexpressionplasmids) to get a sense for how different mutations within a single gene can differ in their effects. Inaddition, a deeper understanding of the nature of the proteins underlying a trait, and the domains of whichthose proteins are composed, may help inferential power in the future. Mutations that are not likely to causeloss-of-function of the whole protein may instead only affect a single protein domain and may have specific,as opposed to general, effects on protein function, interactions and phenotype (RYAN et al. 2013). Whilecurrent tools do not have the ability to predict the effects of such mutations for proteins as a whole (RYANet al. 2013), future advances in functional protein prediction used in conjunction with information from avariety of large-scale assays may enable better mapping of mutations to phenotype.1.5.2 How consistent are genetic interactions across levels of biological diversity?In evolution, genetic interactions (or epistasis) determine the outcome and pace of adaptation, as well asthe potential for genetic reproductive incompatibilities and speciation. From experimental work, we knowthat beneficial mutations often behave non-additively, at least for fitness (e.g., my Chapter 3, BUSKIRKet al. 2017; CHOU et al. 2009, 2011; KRYAZHIMSKIY et al. 2014). Molecular biologists also have a longhistory of studying genetic interactions, traditionally by using suppressor screens. Now, large-scale geneticinteraction studies are being performed that measure epistasis on a genome-wide scale. In organisms whereit is feasible, like S. cerevisiae (COSTANZO et al. 2016) and Schizosaccharomyces pombe (ROGUEV et al.2008), interactions are measured by assessing the growth of double mutant strains. These double mutantsare generated in high-throughput experiments using genetic systems developed specifically for this purpose,allowing for the simultaneous generation and measurement of thousands of strains. In other organisms,like C. elegans (TISCHLER et al. 2008), RNA interference (RNAi) is used to knock down expression of agene, simulating the effect of a null mutation. RNAi can either be used combinatorially (where two genesare simultaneously targeted), or in combination with a homozygous mutant animal. I will refer to theseas double mutants throughout to simplify explanation. Once double mutant organisms have been created,genetic interactions are detected by deviations of the double mutant phenotype from that which is expectedbased on the effects of the two mutations in isolation. The phenotype being measured is usually a proxyfor viability or growth and negative interactions are referred to as either synthetic sick (where the doublemutant is less fit than expected, but still viable) or synthetic lethal (where the double mutant is inviable131.5. What can large-scale datasets tell us about evolution?despite both single mutants being viable). These studies have elucidated some general patterns about geneinteraction networks, such as the existence of ‘hub’ genes, which interact with many more genes than averagein the network, and that functionally related genes can be predicted based on the similarity of their interactionprofiles (DIXON et al. 2009). Can evolutionary biologists take advantage of this wealth of genetic interactiondata to learn about the nature of epistasis in shaping evolutionary trajectories and make predictions aboutpossibly interacting alleles in wild populations or in other species?Large-scale genetic interaction screens have been performed in a few model organisms, but the largestand best-described dataset is for budding yeast, S. cerevisiae (COSTANZO et al. 2016). When large-scaleinteraction studies have been performed in other eukaryotic systems, such as the fission yeast S. pombeand nematode worm C. elegans, the results are often compared with those from S. cerevisiae. In thosecomparisons, the global genetic network properties such as degree of interconnectedness and amount ofcrosstalk between specific biological processes tend to be well-conserved but individual genetic interactionsare less conserved (DIXON et al. 2009; RYAN et al. 2012). In S. pombe, ~17-30% of negative interactionsare conserved with S. cerevisiae in any individual experimental study (when ignoring gene function, seebelow) (DIXON et al. 2008; ROGUEV et al. 2008). DIXON et al. (2008), along with performing a large-scale interaction experiment, also curated the literature at the time for reported S. pombe genetic interactions(note that this set excludes the two previously mentioned experimental studies) and found that 18-23% ofthose interactions were conserved with S. cerevisiae (DIXON et al. 2008). In C. elegans, less than 5% ofinteractions are found to be conserved with S. cerevisiae (BYRNE et al. 2007; TISCHLER et al. 2008).To see how well-conserved interactions might be over multiple species comparisons, I have downloadedthe most recent version of the Biological General Repository for Interaction Datasets (BioGRID) geneticinteraction dataset (release 3.4.151, CHATR-ARYAMONTRI et al. 2017) and chosen the species with themost non-redundant genetic interactions (>1500; C. elegans, D. melanogaster, E. coli, Homo sapiens, S.cerevisiae, and S. pombe) curated from large-scale screens as well as small-scale experiments. For theseorganisms, I found all available pairwise orthologs using Ensembl’s BioMart tool (Ensembl release 90,AKEN et al. 2017, BioMart: KINSELLA et al. 2011), which didn’t include E. coli or S. pombe. I excludedE. coli from analysis but, for S. pombe, I downloaded the set of manually curated orthologs between S.pombe and S. cerevisiae from PomBase (MCDOWALL et al. 2014). For all other species except S. cerevisiae,the fission yeast orthologs were determined by taking the list of budding yeast orthologs with that speciesand finding the fission yeast ortholog for the budding yeast gene. Note that I did not exclude orthologs thatcorresponded to sets of paralogs in the other species, so the total number of orthologous genes considered wasnot equal between both species of a pair. Only genetic interactions were considered for this analysis, but bothpositive and negative interactions were included. For each species pair, I first determined all (non-redundant)pairwise genetic interactions in each species between genes that had orthologs in the other species. I thencompared these lists of interactions between the two species of a pair, checking for overlap where bothinteracting genes were orthologous to interacting genes in the other species. The results are plotted inFig. 1.2.Assuming that interactions are tested randomly with respect to gene identity and to the existence of anortholog in another species, we can determine the proportion of interactions that we expect to be conserved141.5. What can large-scale datasets tell us about evolution?C. elegans0.001 %Index35983(735.6)p<0.0001*363002(2165.1)p<0.0001*Index91314(1385.6)p<0.0001*10528(3419.4)p<0.0001*Index3127(12.3)p<0.005*59179(98.3)p<0.0001*Index22130(14.1)p<0.0001*2998727(53)p<0.0001*D. melanogaster0.005 %Index326112(1059.1)p<0.0001*37847(902.5)p<0.0001*Index1101140(50.3)p<0.0001*10710380(213)p<0.0001*Index3051276(19.9)p<0.0001*328105831(64)p<0.0001*H. sapiens0 %Index17332(26.7)p<0.0001*1711285(304.7)p<0.0001*Index64352(15.1)p<0.0001*66114957(116.1)p<0.0001*S. pombe0.192 %Index489833364(12.2)p<0.0001*4945286973(9)p<0.0001*S. cerevisiae1.204 %Figure 1.2: Conserved genetic interactions between pairs of species. Each pairwise combination of speciesis represented by a single box, found in the corresponding row and column to the species names. Below thespecies name is the percentage of total possible pairwise gene combinations that have been shown to havea significant interaction. The data for each species is coloured by the colour of the species name. Withineach box, for each species, the number of conserved genetic interactions is above the line with the number ofgenetic interactions between genes of that species that have orthologs with the other species below the line.This is followed by the enrichment of conserved interactions observed compared to expected in parenthesesand the p-value of a Binomial test where the null probability is the proportion of total possible gene pairsthat are known to interact in the comparison species. Note that because we included paralogous genes, thenumber of conserved interactions does not necessarily match between the two species in the pair. Below thediagonal is a representation of the phylogenetic relationships between species (not to scale, created usingphyloT: http://phylot.biobyte.de/index.html).151.5. What can large-scale datasets tell us about evolution?between species. For species A, we expect the conserved proportion of its interactions between its genesthat have orthologs in species B to be the proportion of all possible pairs of genes for which a significantinteraction has been found in species B (number of gene pairs tested and determined to interact signifi-cantly/total number of gene pairs = p). Data on the number of non-redundant gene pairs found to have asignificant interaction comes from BioGRID (CHATR-ARYAMONTRI et al. 2017). The total number of genepairs was calculated as the square of the total number of protein-coding genes. We excluded dubious genesfrom these numbers, which gave: C. elegans: 20,222 (release WS260, STEIN et al. 2001), D. melanogaster:13,931 (release R6.17, GRAMATES et al. 2017), H. sapiens: 19,836 (GENCODE release v27, HARROWet al. 2012), S. pombe: 5,064 (release version 30th Jan 2017, MCDOWALL et al. 2014), and S. cerevisiae:5,892 (genome inventory as of 1/18/2017, CHERRY et al. 2011). I used the expected proportions to performBinomial tests to determine whether there were more conserved interactions than expected for each speciesin each species pair. For example, in the C. elegans -D. melanogaster comparison, I first found all significantinteractions between genes in C. elegans that have orthologs in D. melanogaster (a total of n interactions). Ithen determined how many of these interactions were conserved (X conserved interactions where orthologsof both genes interact in D. melanogaster). The Binomial test was then performed using the total number ofinteractions (n) as the number of trials, the number of conserved interactions (X) as the number of successesand the expected proportion of conserved interactions in D. melanogaster (p) as the probability of success.I found that there were significantly more interactions conserved than expected in all cases (Fig. 1.2), evenamong the distantly related species.I find that, when searching for orthologous genetic interactions, there is low conservation between speciesin general. Even in comparisons with S. cerevisiae, where most interactions have been studied, only 14.7 -23.9% of interactions are conserved. When I attempt to account for sampling effort by dividing the fraction ofconserved interactions by the expected proportion based on the total number of significant interactions knownin the other species to get the enrichment of conserved interactions, however, I find that the conservation ismuch higher than what is expected by chance (Fig. 1.2, numbers in parentheses; all Binomial tests aresignificant). Conservation generally shows a pattern of decreasing enrichment with increasing phylogeneticdistance and is very high between the animals (C. elegans,D. melanogaster, andH. sapiens). The enrichmentbetween the yeasts, however, is much lower than the enrichment found between the animals or between eachyeast species and each of the animals. It is possible that interactions found within multicellular animalsare more likely to be conserved with other multicellular animals because of conserved specialization oforthologous genes’ functions in developmentally complex organisms. In the yeasts, however, the genes maybe less specialized and have more redundancy in function, leading to fewer specific (and therefore conserved)genetic interactions.There are some important caveats to this analysis, however, that may lead to systematic overestima-tion of enrichment and underestimation of p-values. If there is a bias towards testing interactions betweenconserved genes (i.e., those most likely to have orthologs in other species) then we may be biased towardsfinding conserved interactions. We know that interactions are not tested randomly, with some studies fo-cussing on particular biological processes (e.g., BYRNE et al. 2007 targeted genes in signal transductionpathways), and it is common practice to test a smaller number of ‘query’ mutations in combination with a161.5. What can large-scale datasets tell us about evolution?larger array of mutations, thus biasing the information gathered in favour of the chosen query genes. Sucha bias could explain the difference in enrichment observed between the animals and the yeasts. Becausefewer interactions have been tested in the animals, they are more biased towards interactions that are morelikely to be conserved. There is a much greater number of interactions that are known in the yeast species,however, with almost all possible interactions having been tested in S. cerevisiae (COSTANZO et al. 2016),although many essential or nearly essential genes remain to be tested. Thus, there is less bias and a numberof conserved interactions that is closer to the expectation. Additionally, we allowed each gene in our anal-ysis to have multiple orthologs in the other species, but we have not accounted for these additional chancesof finding a match in our Binomial test. By restricting our analyses to gene sets that have orthologs in allspecies being compared and that have been tested for interaction in those species, in addition to either usingonly orthologs with a single match in the other species or using probabilities weighted by the homologyrelationships, we could improve our ability to compare conservation between species pairs. Further, morestudies covering larger, and less-biased, portions of the genome will allow for better estimates of the trueenrichment of conserved interactions. An additional complication in using sequence-based homology rela-tionships to assess conservation is that they do not always correspond to functional homology. For example,KACHROO et al. (2017) found that two genes in yeast can be replaced by non-orthologous genes from E. coli.If anything, these types of mismatches between sequence and functional orthology should make estimates ofconservation more conservative when based on sequence data alone.In addition to caveats to the analysis, there are other potential problems in using this kind of data forevolutionary interpretations. Most of the interaction data, and especially data from large-scale interactionstudies, comes from loss-of-function mutations (either gene deletions or RNAi knockdown), which may notrepresent all natural mutations (see Section 1.5.1). In addition, the resulting alleles are generally either mildlydeleterious or neutral. Interactions between deleterious mutations may differ in a biologically significant wayfrom interactions between beneficial mutations. Also, because most large-scale genetic interaction mappingstudies have used proxies for growth rate as the phenotype of interest (DIXON et al. 2009), there is not alot of information about how interactions may differ in kind depending on the phenotype being measured.However, the focus on growth rate enables us to interpret the results in terms of fitness. Because of theseissues and those of incomplete and biased data, the presence of an interaction in curated data may be used asevidence that epistasis is occurring between mutations in two different genes but the absence of interactiondata should not be similarly taken as evidence of no epistasis. These datasets could be especially useful formapping BDM incompatibilities between species where candidate genes could be found that are known tohave synthetic sick or synthetic lethal interactions in a closely related model organism.Despite low conservation of individual connections, larger-scale conservation can be useful in predict-ing functional evolution. Between S. cerevisiae and S. pombe, RYAN et al. (2012) found that, within proteincomplexes, interactions were well conserved (positive interactions: 70%, negative interactions: 68%). In thissame comparison, as the level of biological association between the genes decreased, they found decreas-ing conservation of specific interactions (same biological process, positive: 58%, negative: 38%; separateprocesses, positive: 19%, negative: 15%) (RYAN et al. 2012). Also in comparisons between S. pombe andS. cerevisiae, similar percentages of conserved interactions were found by ROGUEV et al. (2008), and the171.5. What can large-scale datasets tell us about evolution?general patterns were supported by FROST et al. (2012). The frequency of interaction between biologicalprocesses is highly conserved, however, even if the exact connections are not (RYAN et al. 2012). In addi-tion, highly connected hub genes found in C. elegans are also highly connected in other animals, suggestingconservation of certain genes as hubs across species (LEHNER et al. 2006). In light of these results, if weknew the biological processes underlying adaptation to a given environment, we may be able to predict theamount of epistasis that would be found between potentially adaptive alleles in that environment by us-ing data on genetic interactions within and between those processes. This information could indicate howquickly evolution might be expected to proceed as well as whether it might be constrained to certain genetictrajectories. Similarly, we might be able to hypothesize about the amount of negative epistasis expectedto be present between populations adapted to different environments based on the processes important forthose environments. Further developments and increased data in this field will help to inform hypothesesabout specific interacting alleles and determine the general patterns that we can expect to hold true for allorganisms, elucidating how this knowledge can be applied to natural systems.1.5.3 How do genetic interactions depend on the environment?If we want to use large-scale genetic interaction data to help us understand evolution, an important con-sideration is how the observations are dependent upon the environment. Experimental studies are gener-ally performed under a single set of standard laboratory conditions, thus not providing information aboutenvironment-dependence. In Chapter 3, I give an example of epistatic relationships changing with chang-ing concentrations of a single stressor. These kinds of changes may be somewhat predictable, but probablyonly in the special case of changing severity of the evolutionary environment in which the mutations wereselected (see Discussion). Another study on reproductive isolation in yeast (HOU et al. 2015) tested manyintraspecific crosses on different culture conditions. All offspring were chosen to have high viability in therich lab medium (YPD), but the authors found that the hybrid viability was environment-specific and variedamong crosses. These results indicated that the underlying genetic incompatibilities were dependent uponthe test conditions. Other evolution-based studies provide case studies showing that epistasis can dependon the environment in which it is measured (e.g., REMOLD and LENSKI 2004; WANG et al. 2009), but fewdirected investigations of the large-scale generality of the phenomenon have been performed.A few studies have attempted to address this issue, primarily in S. cerevisiae. JASNOS et al. (2008)measured the maximum growth rate of single and double gene deletion strains in a benign environmentand several stressful environments, finding that epistasis became positive (alleviating), on average, in morestressful environments. They attributed this change to the general properties of decreasing growth rate instressful environments; when growth rate is impaired, other defects caused by mutations have less of animpact. They did not focus on how individual interactions changed with the environment, however, orwhether individual interactions changed in a predictable way. While alleviation of deleterious effects maybe the overall pattern, it is possible that individual interactions will differ, potentially in an important way. Inaddition, JASNOS et al. (2008) specifically chose to investigate deleterious mutations, but it is possible thatneutral or beneficial mutations could show altogether different patterns. In addition, both targeted (ST ONGEet al. 2007) and broad (BANDYOPADHYAY et al. 2010) surveys of genetic interactomes in the presence and181.5. What can large-scale datasets tell us about evolution?absence of DNA-damaging agent methyl methanesulfonate (MMS) in yeast indicate that interactions can beenvironment-dependent. These studies indicate that changes in interactions may be somewhat predictablebecause the genes that had many changes in their interactions were often ones that were known to be sensitiveto MMS when knocked out or genes with a role in DNA repair (BANDYOPADHYAY et al. 2010), and similarresults were found in E. coli (KUMAR et al. 2016). Again in yeast, deletion mutations of paralogs from awhole genome duplication were also found to have altered patterns of epistasis under different experimentalconditions (MUSSO et al. 2008). Paralog double mutants that were sensitive to certain stressful conditionswere generally found to have functions related to that condition (MUSSO et al. 2008). Finally, condition-dependent epistasis is also predicted from models of metabolic networks in yeast (BARKER et al. 2015;HARRISON et al. 2007) and other microbes (JOSHI and PRASAD 2014), and a small number of the predictedinteractions in yeast from HARRISON et al. (2007) were verified in vivo.Further information is needed in this field, especially from experimental data. It would be relativelysimple and worthwhile to test double mutant strains made in rich medium conditions in a variety of otherconditions, as long as the mutants are viable in the original conditions. One generally ecologically-relevanttrait that would be useful to test is temperature-dependence. There is a lot of interest in the effects oftemperature right now as it relates to climate change and varies both geographically and temporally. Forlab-reared organisms, testing for the effect of temperature on a mutant phenotype could be as simple asrearing the organism in a temperature-controlled incubator. I would generally expect growth to worsenin temperatures that are increasingly divergent from the optimum (due to the instability of the underlyingproteins or slowing of enzymatic reactions), but especially large changes in phenotype could be examined todetermine the underlying causes. From these kinds of experiments, I would hope to find out whether networkproperties remain stable, in general, and whether changes in environment have predictable and/or consistenteffects on epistasis. These results would again have implications for how adaptation might proceed after anenvironmental change.Changing epistasis in changing environmental conditions also has implications for how we conceptual-ize reproductive isolation in speciation. Reproductive isolation builds up from a combination of isolatingbarriers, which have been traditionally categorized as either extrinsic or intrinsic. Extrinsic barriers arise asa mismatch of the organism’s phenotype to the environment in which it is found. Common examples includeimmigrant inviability and hybrid partial inviability due to intermediacy of phenotype in one of the selectiveenvironments. Genetic incompatibilities, on the other hand, have been traditionally classified as intrinsicbarriers, independent of the environment. If the incompatibility of alleles often depends on the environmentin which it is measured, then these ‘intrinsic’ isolating factors will have an extrinsic basis and we will haveto include these kinds of cases in our models of speciation. If hybrids between species are likely to findthemselves in qualitatively different environments from their parents, knowing how epistasis changes withenvironmental conditions will be especially important for the stability of a reproductive barrier.191.6. Summary1.6 SummaryDespite the many recent advances in our knowledge about the genetic basis of evolution, there is still muchto be discovered. Experimental evolution using model organisms is helping to lead the way in the moleculardissection of evolution, informing us about the repeatability of evolution under different circumstances, andhow that repeatability depends on the genetic properties of the organism and their capacity for adaptation.In addition, we have gathered evidence on the profound effects that epistasis can have on both adaptationand speciation. These insights have allowed us to begin to interpret the abundant genomic data accumulatingin natural systems, but I believe that there is still more to be gained by utilizing knowledge acquired fromlarge-scale molecular studies of model organisms. These studies have informed us about the nature of manygenes in a few model organisms, and how these genes are organized into complexes, pathways and differentlevels of biological processes. Further, these datasets have uncovered general properties of genetic interac-tion networks, and which of these are well-conserved among species. Unfortunately, these large datasets areoften collected using gene knockout mutants, or other loss of function mutations, and these mutations tendto be deleterious. Relatively little is known about how natural mutations or beneficial mutations might differ,and we might expect these types of mutations to be categorically different in their properties and epistaticrelationships. In addition, there is still relatively little known about how epistatic relationships change withchanging environments, especially for mutations that were beneficial in the original environment. By con-sidering the data produced by large-scale studies when making evolutionary hypotheses and utilizing thebiological tools developed by molecular biologists to perform directed experiments when the necessary in-formation is lacking, I believe that we can greatly improve our ability to interpret evolutionary genetic dataand our predictions for evolutionary trajectories.20Chapter 2Too much of a good thing: The unique andrepeated paths toward copper adaptation2.1 IntroductionIn his book, Wonderful Life (GOULD 1989, p. 51), Stephen J. Gould famously opined that evolution is ahistorical and contingent process, so much so that “any replay of the tape would lead evolution down a path-way radically different from the road actually taken.” While this is undoubtedly true when one considers thefull complexity of an organism, refrains are often observed in evolution at the trait level. Repeated evolu-tion, defined as ‘the independent appearance of similar phenotypic traits in distinct evolutionary lineages’(GOMPEL and PRUD’HOMME 2009) has been documented in both ecological and clinical environments atall taxonomic levels, e.g., repeated loss of stickleback lateral plates in freshwater (SCHLUTER et al. 2004),ecomorphs of Anolis lizards (LOSOS 1992), the acquisition of “cystic fibrosis lung” phenotypes in Pseu-domonas aeruginosa in cystic fibrosis patients (HUSE et al. 2010), to name but a few. The developmentof sequencing technologies has recently allowed biologists to ask whether parallel genetic changes underlieobservations of parallel phenotypic change. In some cases, parallel phenotypic evolution has been attributedto parallel genotypic evolution, for example, repeated changes to cis-regulatory regions of the same gene—the pigmentation gene yellow—underlie changes in wing pigmentation in male Drosophila (PRUD’HOMMEet al. 2006). At the other extreme are cases where different genetic targets underlie similar phenotypicshifts; for example, yeast adapting to rich media converged in fitness via a variety of genetic mechanisms(KRYAZHIMSKIY et al. 2014), and beach mice adapting to sandy coastal dunes from the Gulf and Atlanticcoasts of Florida converged in coat coloration via different mutations (MANCEAU et al. 2010). In such cases,unique evolutionary trajectories at the genetic level appear repeatable at the phenotypic level.The degree of phenotypic repeatability is inherently linked with the genomic target size of appropriatemutations, with single-locus Mendelian traits with fewer target sites (and hence higher repeatability) at oneextreme and quantitative traits at the other extreme. Even when multiple genes underlie a selected trait,however, there may be relatively few sites that, when mutated, have the magnitude of effect and sufficientlyminor deleterious side effects to improve fitness overall (STERN 2013). Such pleiotropic constraints arethought to explain why cis-regulatory sites more often contribute to adaptation than trans-regulatory changes(GOMPEL et al. 2005; STERN 2000). The size of the population and the manner in which it reproduces arealso critical. Large populations have access to rarer mutations, particularly those of large effect (BURCH andCHAO 1999), increasing the chance that the best of these mutations will fix in independent evolutionary trials(BELL and COLLINS 2008). Mutations with particularly high fitness are also more likely to fix in asexual212.1. Introductionpopulations, because clonal interference reduces the chance that minor-effect mutations establish (ROZENet al. 2002), unless adaptive mutations are so common that coalitions of mutations establish together (FOGLEet al. 2008; LANG et al. 2013).The nature and severity of environmental challenge will also affect the degree of repeatability at both thegenotypic and phenotypic levels. If the environmental change is so severe that the population cannot replaceitself and there are only a small fraction of mutations whose benefits are large enough to bring absolutefitness above one (BELL and COLLINS 2008), then adaptation would be more repeatable. On the otherhand, if an organism is adapting via mutations whose effects are small relative to the distance to the fitnessoptimum, nearly half of mutations are predicted to be beneficial (FISHER 1930), and adaptation would be lessrepeatable. The genomic target size must also depend on the nature of mutations required: when adaptationcan be accomplished by the loss of a function, adaptive mutations can potentially arise in any step alongthe pathway leading to that function via a variety of mechanisms (e.g., single base pair changes leading topremature stop codons early within a gene, movement of transposable elements within a gene, mutations inthe promoter that alter transcription factor binding sites, etc.). In contrast, if the environmental challengerequires the appearance of a novel trait, or an alteration of an existing trait, the number of genomic targetsis likely diminished. Despite these long-standing theoretical predictions, empirical data have only recentlybeen catching up, largely due to breakthroughs in sequencing technology (e.g, SCHNEEBERGER 2014).In this study, we set out to determine the repeatability of adaptive evolution at the genotypic and phe-notypic levels using short-term experiments with the yeast, Saccharomyces cerevisiae. We purposefullyemployed a short-term experimental design in an attempt to avoid the potential influence of epistasis limit-ing the mutations that are sampled (CHOU et al. 2011; KVITEK and SHERLOCK 2011). The design of theexperiment was similar to a previous study in our group where we examined adaptation to the fungicidenystatin (GERSTEIN et al. 2012). In both cases, multiple isogenic lines of yeast were exposed to inhibitorylevels of either nystatin or copper, with levels chosen to be slightly higher than those in which growth oc-curred reliably. Lines that showed growth were isolated and analyzed. Through whole-genome sequencingof 35 lines that evolved tolerance in the nystatin experiment, we found that adaptation repeatedly involvedthe same four genes in a single pathway leading to the production of ergosterol (GERSTEIN et al. 2012),the membrane-bound target of nystatin (WOODS 1971). Indeed, of the 20 unique mutations identified, 18involved the same two genes (11 different sites in ERG3 and 7 in ERG6). In hindsight, the highly repeatednature of this adaptation may well be explained by the narrowness of the environmental challenge: the cellscan survive by blocking the production of ergosterol, and this can be accomplished through loss-of-functionmutations in the ergosterol biosynthesis pathway (particularly in ERG3 or ERG6). We thus set out to as-sess the degree of repeatability in the face of an entirely different environmental challenge: high copperconcentrations, where loss-of-function mutations are less expected.Copper is a micro-nutrient that is essential for several different enzymatic processes in yeast (cytochromeoxidase involved in respiration, superoxide dismutase involved in defense against oxidative damage, and theFet3p ferro-oxidase involved in iron uptake, GRADEN and WINGE 1997). Thus, unlike nystatin, cells can-not entirely block copper uptake. On the other hand, copper is extremely toxic at high concentrations, bothbecause it displaces other metal co-factors from proteins and because it produces highly reactive oxygen222.2. Materials and Methodsspecies, including damaging hydroxyl radicals (PEÑA et al. 1999). The fact that multiple cellular processesrequire copper, that multiple cellular compartments are involved in copper sequestration (especially vacuolesand mitochondria), and that multiple processes are impacted negatively by copper (PEÑA et al. 1999) sug-gests that adaptation to high copper concentrations may occur through a variety of mechanisms. Here wereport the results of a short-term adaptation experiment to this toxic but essential metal. Through wholegenome sequencing, we identify the nature of the genetic changes that underlay the evolutionary rescue of34 lines of S. cerevisiae exposed to inhibitory copper concentrations.2.2 Materials and Methods2.2.1 Evolution of haploid mutation linesMutations were acquired in haploid lines of the common lab strain, BY4741 (MATa his341 leu240 met1540ura340), obtained from Open Biosystems in 2009. Preliminary experiments determined that BY4741 grownin liquid YPD + 12.5mM CuSO4 does not show consistent growth, but that some populations begin growingat different times, a stochastic pattern of growth we have previously shown to be consistent with benefi-cial mutations in other environments (GERSTEIN et al. 2012). To initiate mutation acquisition, we streakedBY4741 from frozen onto a YPD plate and randomly chose a single colony to grow overnight in 10mLYPD, shaking at 30C. We added 4.5mL of this common wild type stock to 185.5mL YPD + 12.5mMCuSO4 (hereafter referred to as ‘copper12 medium’). We placed 1mL aliquots into 180 inner wells of three96 deep-well boxes, with 1mL of dH2O in the outer wells. Inner wells were used, with dH2O in the outerwells, in an effort to reduce evaporation. Boxes were maintained shaking on a platform shaker at 30C. Allboxes were checked daily by visual examination of the bottom of the wells. Growth was recorded whenwe saw precipitate on the bottom of a well and was first observed after 7 days of incubation (Table A.1).Twenty-four hours after growth was first seen, we manually mixed the well and froze 500µL of culture in15% glycerol. In this way we isolated 56 ‘putative mutation lines’ within 14 days, post-inoculation.At the end of the mutation-accumulation phase we struck each putative mutation line from the freezerstock onto a single YPD plate and grew them at 30C. Two of the putative mutation lines did not grow within72 hours and were excluded from this point forward. Fourteen lines exhibited very small colonies, typicalof petite colonies that have lost mitochondrial function. One of our initial goals in acquiring these mutationlines was to measure their fitness in heterozygous form; to avoid assaying non-nuclear mutations, these lineswere also excluded from the set of lines we genotyped and phenotyped. From each remaining line we hap-hazardly chose eight colonies and inoculated each colony into 1mL copper12 medium and 1mL YPD (byusing the same pipette tip) and incubated them at 30C with shaking. In six cases, none of the eight coloniesgrew in copper12 medium within 72 hours, leaving us with 34 copper-adapted mutation lines (CBM: ‘Cop-per beneficial mutation’ lines, Table 2.1). From the paired YPD culture descended from the same colony(limiting exposure of our stocks to copper), 500µL was added to 500µL 30% glycerol and frozen.232.2.MaterialsandMethodsTable 2.1: Mutations identified in the CBM lines. CUP1 coverage for each line is provided in the second column and does not account for additionalcopies via chrVIII aneuploidy.CUP1 Genome Position Mutation Position Amino acidCBM line coverage (chr.bp) Gene (Watson strand) (from 5’ end) change ExchangeabilityCBM1 1.61 X.412600 VTC4 C>T 800 Trp>StopXI.105507 FAS1 G>T 4837 Val>Phe 0.207XVI.420661 intergenic A>TCBM2 2.00 chrII aneuploidyCBM3 2.48 VII.150650 intergenic G>TchrII aneuploidyCBM4 3.26 mito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM5 3.78 X.413174 VTC4 C>A 226 Glu>StopX.654261 intergenicc T>CXIV.284255 intergenic T>GCBM6 3.69 III.100061 BUD3 G>A 3781 Gly>Arg 0.178IV.319466 VAM6 T>A 655 Lys>Stopmito.59168 21S_RRNA A>G 1160 Lys>Arg 0.440mito.69322 tRNA-Arg C>G 34 Arg>Gly 0.251CBM7 0.91 II.365359 TRM7 C>T 361 Val>Ile 0.537III.306327 intergenic G>TIV.143017 YDL176W G>T 921 Ser>SerIV.177435 CLB3 T>G 663 Thr>ThrV.392908 BOI2 C>T 805 Glu>Lys 0.323VII.949946 SMI1 C>A 954 Lys>Asn 0.457IX.370383 intergenic C>GXV.215888 MAM3 C>G 250 Gly>Arg 0.178chrVIII aneuploidychrXVI aneuploidyCBM11 2.46 X.413020 VTC4 1D indel (GG A/- AA) 380 Phe>Ser+frameshiftXI.566200 CCP1 A>G 999 Phe>PheXII.605283 intergenic 1D indel (TT A/- AA)chrII aneuploidyCBM13 4.02 X.412247 VTC4 C>A 1153 Glu>StopX.654261 intergenicc T>CCBM14 2.15 XV.215018 MAM3 C>T 1120 Val>Ile 0.537CBM16 0.28 VII.480836 PMA1 A>T 1831 Phe>Ile 0.181chrII aneuploidyContinued on next page242.2.MaterialsandMethodsTable 2.1 – continued from previous pageCUP1 Genome Position Mutation Position Amino acidCBM line coverage (chr.bp) Gene (Watson strand) (from 5’ end) change ExchangeabilityCBM17 0.98 X.412325 VTC4 A>G 1075 Tyr>His 0.197XIII.711207 ESC1 C>T 4075 Leu>Phe 0.336XIII.821262 FCP1 T>C 1007 Leu>Ser 0.212chrVIII aneuploidychrXVI aneuploidyCBM18 2.80 V.303094 VTC1 G>T 289 Asp>Tyr 0.227VII.548326 GSC2 C>T 63 Asp>AspXI.646356-onwards FLO10d A>GCBM20 1.82 VII.480463 PMA1e G>T 2204 Ala>Asp 0.193XV.215332 MAM3 C>T 806 Ser>Asn 0.390XVI.84024 YPL247C C>T 173 Gly>Asp 0.188chrII aneuploidyCBM21 1.12 VII.971165 PFK1 G>C 2570 Pro>Arg 0.254X.654261 intergenicc T>CchrII aneuploidychrIII aneuploidychrVIII aneuploidyCBM22 0.78 V.302818 VTC1 1D indel (CA C/- CA) 13 Pro>His+frameshiftchrVIII aneuploidychrXVI aneuploidyCBM24 0.77 IV.805485 intergenic A>GIV.805517 intergenic G>ACBM25 2.28 IV.530697-onwards ENA5f A>GIX.621992 MLP1 G>T 2188 Glu>StopCBM26 0.66 VII.480470 PMA1 T>G 2197 Thr>Pro 0.164chrI aneuploidychrV aneuploidychrVIII aneuploidyCBM29 1.08 VII.1376 intergenic A>GVII.480463 PMA1e G>T 2204 Ala>Asp 0.193XV.566240 intergenic G>CchrII aneuploidyCBM30 3.30 chrII aneuploidyCBM33 2.40 VII.618173 VHT1 G>C 1686 Ile>Met 0.279VIII.321332 SBE22 A>T 919 Met>Leu 0.513X.412080 VTC4g C>T 1320 Trp>StopContinued on next page252.2.MaterialsandMethodsTable 2.1 – continued from previous pageCUP1 Genome Position Mutation Position Amino acidCBM line coverage (chr.bp) Gene (Watson strand) (from 5’ end) change Exchangeabilitymito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM34 2.92 X.412080 VTC4g C>T 1320 Trp>StopXI.364518 intergenic complex 1I indel (GA>AAT)mito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM36 2.15 X.412080 VTC4g C>T 1320 Trp>Stopmito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM37 2.04 X.412080 VTC4g C>T 1320 Trp>Stopmito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM44h 1.39 X.412080 VTC4g C>T 1320 Trp>Stopmito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM45 2.68 X.412080 VTC4g C>T 1320 Trp>Stopmito.24277a COX1b 1D indel (GG C/- CC) 10460 intronCBM46 2.74 X.412643 VTC4 T>A 757 Arg>StopXI.438478 DID4 A>G 701 Gln>Arg 0.366CBM47 2.34 V.302909 VTC1 C>A 104 Ser>StopCBM49 3.14 V.438349 intergenic G>CXII.1034221 HMG2 C>T 1595 Pro>Leu 0.258XIII.420239 intergenic A>CXIV.265933 GCR2 T>A 598 Lys>StopCBM51 2.50 II.444465 FES1 C>T 229 Asp>Asn 0.201IV.310552 intergenic A>GCBM53 2.88 V.180433 PRP22 C>T 1593 Ile>IleCBM54 1.98 VII.1077964 MAL12 G>T 1366 Gly>StopCBM55 2.25 (no mutations except to CUP1)a This mutation falls in an intron of COX1 but causes a frameshift in an overlapping predicted gene, A15_Beta.b Identical COX1 mutation observed in seven different lines.c Identical intergenic mutation observed in three different lines.d The alignment formed a 100% match to the beginning of FLO10 until XI.647464, at which point the alignment switched to a perfect match to a similar region downstream, starting at XI.648031. Insilico qPCR confirmed the absence of unique intervening sites (CACCAGCTCTTCCTGGTCGT and CACCAGCTCTTCCTGGTCGT) within the FASTQ files for CBM18 (but present in other CBM lines),indicating a deletion in this region.e Identical PMA1 mutation observed in two different lines.f The alignment within this region exhibited a 100% match to the beginning of ENA5 but switched to a 100% match to ENA1 from approximately site IV.527743, suggesting a deletion. Because of thehighly repetitive nature of this array, in silico qPCR was unable to uniquely identify the missing positions.g Identical VTC4 mutation observed in six different lines.h CBM44 was sequenced from the original population, not the representative colony.262.2. Materials and Methods2.2.2 Sequencing of haploid mutation linesFreezer culture from each CBM line was streaked onto YPD plates and grown for 48 hours at 30C. Wehaphazardly picked a single colony for each line and grew it for 24 hours in 50mL of YPD at 30C withshaking. Genomic DNA was extracted using standard protocols (SAMBROOK and RUSSELL 2001). Proto-cols supplied by Illumina were followed to create barcoded libraries for each line (2011 Illumina, Inc., allrights reserved). We sequenced 100bp single-end fragments for each line, pooling 12 uniquely barcodedstrains in each lane on an Illumina HighSeq 2000. Twelve samples were rerun to obtain sufficient depth ofcoverage using 100bp paired-end fragments: CBM18, 20, 21, 22, 24, 25, 26, 29, 30, 34, 36, 44.The resulting genomic sequence data was processed using Illumina’s CASAVA-1.8.0 as in GERSTEINet al. (2012). We called SNPs and small insertions and deletions using configure-Build.pl and parsed theoutput files with custom UNIX and perl scripts. We took advantage of Illumina data from the previous setof experiments with nystatin (GERSTEIN et al. 2012), which were initiated from the same BY4741 cul-ture, to determine the mutations that are common to our strain background yet different from the S288Creference genome (scergenome.fasta downloaded from the Saccharomyces Genome Database, http://downloads.yeastgenome.org/genome_release/r64/); all such differences were removed fromthe dataset. Given that our lines were haploid, mutations called as heterozygous were discarded (likely align-ment errors), as were SNP and indel calls of low quality (Q < 20). Remaining variants were checked inthe alignments, using tview in samtools-0.1.7a (LI et al. 2009). SNPs were independently called using thebwa software package to perform the alignment along with samtools-0.1.7a to identify SNPs, using the -bq1 option to limit data to reliable alignments (LI et al. 2009), confirming all SNPs found by CASAVA (Table2.1).To assess chromosomal aneuploidy events, the total depth of coverage for each chromosome was calcu-lated as the proportion of sequenced sites mapping to a particular chromosome, relative to the proportion ofknown mapped sites located on that chromosome within the yeast reference genome (as reported by config-ureBuild.pl in Illumina’s CASAVA-1.8.0 package).Intergenic mutations were analyzed for gains and losses of predicted TF binding sites using Cis-BP,a tool offered by the online Catalog of Direct and Inferred Sequence Binding Preferences (available athttp://cisbp.ccbr.utoronto.ca/TFTools.php). Cis-BP compares two sequences (i.e., onewildtype and one mutant allele) for differential transcription factor binding inferred based on the rela-tionship between similarity in DNA binding domain amino acid sequence and DNA sequence preferences(WEIRAUCH et al. 2014).2.2.3 Expected frequency of mutations causing non-synonymous and stop codonsThe expected frequency of mutations that would generate a particular type of amino acid change (synony-mous, non-synonymous, or stop) was calculated from the observed codon frequency in S. cerevisiae (http://downloads.yeastgenome.org/unpublished_data/codon/ysc.gene.cod; produced byJ. Michael Cherry based on the 6216 ORFs within the “Saccharomyces Genome Database” (SGD) as of Jan-uary 1999). For each position in each codon, the frequency of all possible mutations was calculated according272.2. Materials and Methodsto the observed spectrum of mutations reported by LYNCH et al. (2008) based on previous studies in yeast.(Similar results were obtained using other mutation spectra, including a uniform distribution, the spectrumobserved by LYNCH et al. (2008) in their mutation-accumulation study, and the observed mutation spectrumin this study.)Summing over the whole genome, the expected frequency of mutations leading to stop codons is 5.78%.The expected frequency of non-synonymous mutations is 73.0% among all possible codon changes or 77.6%among only the synonymous and non-synonymous changes (excluding those going to or from a stop codon).The expected frequency of mutations causing any change to the amino acid sequence is 78.9%, which issimilar to the expectation used previously (78.7%) based on a uniform frequency of mutations (WENGERet al. 2011). Because mutations are biased towards transitions and away from G/C, we recommend using theestimates reported here, which are based on the greatest amount of data regarding the mutational spectrum(LYNCH et al. 2008).2.2.4 Determination of CUP1 copy numberUsing samtools-0.1.7a, the alignments of all CBM lines were manually checked at genes that are known orsuspected to be important for acclimation to high levels of copper in S. cerevisiae, with a particular focus ongenes that were previously identified to be up-regulated under high levels of copper: BSD2, CCC2, COX23,CTR2, CUP1-1, CUP1-2, CUP2, FET3, FMP23, GEF1, HAA1, PCA1, SCO1, SCO2, SLF1, VMA3. Thealignments were normal for all of these genes (including 500bp up and downstream), except CUP1-1 andCUP1-2 on chromosome VIII. In this region, large gaps were consistently found spanning the duplicatedcopies of these genes, caused by alignment ambiguities in this tandem repeat region.To measure CUP1 copy number without having to rely on alignments, we carried out the bioinformaticsequivalent of a qPCR analysis (in silico qPCR; GERSTEIN et al. 2014) by using the unix command “grep”to directly count the number of fastq fragments containing “primers” in the CUP1 region. Specifically, wesummed the number of fragments containing the 16 bp fragment from the very beginning and from thevery end of CUP1, plus two 16 bp fragments between CUP1-1 and CUP1-2 (TTTCAAGAGAACATTT andGGGTGGTGAAGTAATA), searching for all four in the forward and reverse directions (e.g., using “zgrepTTTCAAGAGAACATTT *fastq*”). We then repeated this in silico qPCR procedure for three unique geneson chromosome VIII as controls (using the first 16 bp ofDED81,DUR3, RIX1 in both the forward and reversedirection). A BLAST search was used to confirm that these fragments aligned only to the appropriate genes(http://www.yeastgenome.org/cgi-bin/blast-sgd.pl). We also conducted this procedurewith the 35 BMN lines isolated in nystatin (‘beneficial mutation nystatin’; GERSTEIN et al. (2012)), whichwe initiated from the same ancestral genetic background, providing a baseline for comparison. Relative tothe three control genes, the BMN lines had an average of 18.13 copies of CUP1 (range: 12.40 – 30.45). Notethat although the S288C reference genome on the Saccharomyces Genome Database (SGD) reports only twoCUP1 copies, an isolate of S288C was recently found to contain about 14 copies by Southern analysis (ZHAOet al. 2014). Our data are thus consistent with our ancestral BY4741 strain having undergone amplificationin this region, and we report the number of CUP1 copies in our copper adaptation strains relative to the282.2. Materials and MethodsTable 2.2: Oligonucleotides employed for real time PCR (RT), Southern blot analysis (SB), and genotyping(GT) in the forward (F) and reverse (R) directions.Primer name Sequence ExperimentCUP1-F AGCTGCAAAAATAATGAACAATGC RTCUP1-R GCATTTGTCGTCGCTGTTACA RTTAF10-F AAGTTGTTCTGACGGTGAACGA RTTAF10-R GCGACCTATATTGAGCCCGTATT RTCUP1-F 5Biosg/TTAATTAACTTCCAAAATGAAGGTCA SBCUP1-R 5Biosg/AGACTATTCGTTTCATTTCCCAGAG SBMAM3-F AATGAGTGCCGATACCATCC GTMAM3-R GATTCGTCCCAATCTTTTGC GTVTC4-F GTTCATGATCTAGCAAAGTTTTCG GTVTC4-R GGTAACCAAAATGGGATTGAA GTLYS2-F TCAAGGGCTGAAAAGACAATCAA GTLYS2-R CGACGCAAAGAGATGAAACCA GTaverage across the BMN lines.To test whether levels of CUP1 inferred from in silico qPCR were consistent with levels of CUP1 tran-scription, we assayed RNA levels using quantitative real-time PCR (qPCR). Detailed methods are providedin Section A.1.1. Briefly, we chose 10 CBM lines that spanned the range of CUP1 copy number. A singlecolony of each CBM line and two colonies of BY4741 were inoculated into 1mL YPD + 5.5mM CuSO4(a lower concentration was used to allow growth of all lines, including BY4741) and grown for 12 hours at30C with shaking, at which point RNA was isolated using the RNEasy Mini Kit from Qiagen, following theyeast protocol. Oligonucleotides for qPCR (Table 2.2) were designed using Primer Express (ABI). mRNAlevels of TAF10 were used for normalization, because TAF10 exhibits stable expression across strains andconditions (TESTE et al. 2009).2.2.5 Phenotypic assays of CBM linesTo determine the extent of copper tolerance acquired, we conducted dose-response experiments in deep-wellboxes. Each CBM line was struck from frozen onto YPD and grown for 48 hours at 30C. A single colonywas then haphazardly chosen from each line and inoculated into 10mL YPD, shaking overnight at 30C. Theoptical density of all lines was standardized to the least dense line and 200µL of standardized culture wasadded to 400µL YPD; 15µL was then inoculated into 1mL of 8 different levels of YPD + CuSO4 (0mM,4mM, 8mM, 9mM, 10mM, 11mM, 12mM, 14mM). Four replicates were grown for each line in each level ofcopper. Boxes were maintained shaking on a bench top shaker at 30C. After 72 hours we manually mixedeach well and the optical density (OD) of 200µL of culture was measured on a BioTek plate reader. With thisdata, we determined the IC50 (half-maximal inhibitory concentration) of copper using a maximum likelihoodfitting procedure, as previously described (GERSTEIN et al. 2012).292.2. Materials and MethodsTo assess whether there was a correlation between the ability to grow in elevated levels of copper andfitness in other environments, we conducted a series of growth rate experiments using the Bioscreen CMicrobiological Workstation (Thermo Labsystems) to automate OD readings. From the rise in OD, growthrates were estimated under multiple environmental conditions: YPD + 8mM CuSO4 (‘copper8’); YPD,a standard laboratory rich medium; YPG, a medium that requires yeast cells to respire; and YPD + iron(ferric citrate). The latter environment was of particular interest because of copper’s role in iron uptake viathe Fet3p ferro-oxidase, so growth was assayed at three levels of ferric citrate: 10mM, 40mM and 60mM;we only present the 40mM results in the main text because results were highly correlated across the ironconcentrations (Table A.2). Copper (0.2M Cu(II)SO4·5H2O) and iron (1M C6H5FeO7) stocks were madein distilled water. Iron stock was made at least three days prior to use with occasional vortexing and mildheating to keep the ferric citrate in solution. In both cases, copper or iron stock was added after YPD wasautoclaved, roughly one hour before the addition of yeast culture.Each growth rate assay was initiated in a similar manner to the IC50 assays. Cultures from BY4741 andall CBM lines were struck from frozen and grown on YPD plates incubated at 30C for 2-3 days. Four orfive colonies from BY4741 and a single colony from each CBM line was then inoculated into 10mL YPD,shaking overnight at 30C. Optical density from overnight culture was standardized, and a 1:101 dilution wasconducted into the appropriate medium. For each line, five random wells spanning two 100-well honeycombplates were filled with 150µL of diluted culture. Plates were incubated at 30C with maximum shaking for24 hours on a Bioscreen C, with automated OD readings every 30 minutes. From the raw data, we extractedthe maximum growth rate using a non-parametric spline fit performed by a custom R script (GERSTEINet al. 2012). The maximum growth rates from the replicates of each CBM line were statistically comparedagainst all replicates initiated from BY4741 using a t-test (replicates involving separate wells from a singleBioscreen C plate).2.2.6 Copper tolerance of deletion linesTo assess whether intragenic mutations that arose within our CBM lines are phenotypically similar to knock-out mutations, we measured copper tolerance (IC50) of 21 gene deletion lines (GIAEVER et al. 2002), repre-senting all of the available knockouts for the characterized genes that had mutated in our study (excluding theuncharacterized YPL247C and YDL176W). BY4741 is the progenitor of both the deletion collection and ourancestral strain background, allowing a direct comparison of the impact of deleting these genes. Tolerance(IC50) was determined as above from OD measurements taken across an array of copper concentrations at24 hours. Tolerance assays were conducted in the Bioscreen C and replicated twice, running simultaneouslyon two different machines.2.2.7 Tetrad dissections to isolate single mutationsTo separate the effects of single mutations from other mutations present in the evolved lines (includingextra copies of CUP1), we crossed all of the CBM lines with BY4739, which has a common genotype302.2. Materials and Methodsyet opposite mating type and different auxotrophies than BY4741, the progenitor of our lines. We thenattempted to sporulate the resulting diploid lines, focusing on a subset that contained each common mutationor aneuploidy and the fewest number of additional mutations (⇠1/3 of the lines). Detailed methods areprovided in Section A.1.2.We encountered substantial difficulties in obtaining tetrads from our strains; BY4741, a derivative ofS288c, is known to be a poor sporulator (BEN-ARI et al. 2006; DEUTSCHBAUER and DAVIS 2005). Inparticular, despite many attempts, no tetrads were obtained for CBM16 (PMA1 mutation plus chrII aneu-ploidy), CBM26 (PMA1 mutation plus chrI, chrV and chrVIII aneuploidy), CBM29 (PMA1 mutation pluschrII aneuploidy), CBM47 (VTC1 mutation), or CBM55 (no mutation identified other than extra copies ofCUP1).We were able to sporulate CBM2 (chrII aneuploidy), CBM14 (MAM3 mutation), CBM25 (MLP1 andENA5 mutations), and CBM34 (VTC4 mutation). CBM25 was not initially chosen for tetrad dissection butwas dissected as a contaminate of CBM22 (VTC1 plus chrVIII and chrXVI aneuploidy), as detected bysubsequent sequencing. CBM25 contaminating cells were likely positively selected during the sporulationprocedure given that the aneuploid lines in our experiment, like CBM22, had very low sporulation rates.The genotype of resulting spores was then determined (see Section A.1.2; PCR primer information inTable 2.2). In brief, for CBM14 and CBM34 tetrad lines, MAM3 and VTC4, respectively, were amplifiedby PCR. All SNPs showed the expected 2:2 segregation pattern in the four spores of each dissected tetrad.CBM25 spores were sequenced on Illumina HiSeq 2000, which is when the strain was discovered to beCBM25 (bearing a mutation in MLP1 and ENA5), not CBM22. The segregation pattern for the additionalcopy of chrII in CBM2 spores was determined by the segregation patterns of LYS2 alleles. To quantify thesegregation patterns of CUP1 among the spores, Southern blots with CUP1 specific probes were performed.We isolated genomic DNA and ran a Southern blot on three separate occasions for each spore. Band inten-sity was quantified in ImageJ (ABRAMOFF et al. 2004) using the “background corrected density" macro toestimate CUP1 copy number.2.2.8 Fitness effect of single mutations on growth rate and copper toleranceTo measure the fitness effects of the mutations isolated by tetrad dissection, growth rate assays were con-ducted within the Bioscreen C using either YPD + 9mM CuSO4 (‘copper9’) or YPD, as described abovewith the following exceptions. Yeast was occasionally taken from a lawn plated from frozen cells rather thanfrom single colonies (the sporulated lines had been bottlenecked to a single colony just prior to freezing,and so a second bottleneck at this stage was less essential). Optical density was not standardized for thecopper9 experiments as this was deemed to have little effect on inferred growth rates. For each line, two(copper9) or four (YPD) non-adjacent wells were filled with 150µL of diluted culture and allowed to growfor 24 hours. This procedure was performed three times in the copper9 environment, and once in YPD todetermine whether these lines were affected in their ability to grow in the nutrient-rich environment. Themean maximum growth rate was determined for each Bioscreen C assay in the copper9 environment, andstatistics were performed using these means as data points.312.3. ResultsCopper tolerance was determined for a subset of spores through dose-response experiments as describedfor the knock-out lines except that two separate Bioscreen C runs were performed, with two replicate wellsper run (Figure A.1). Specifically, we assayed IC50 for two spores (among all of the tetrads for each line)that carried mutations of interest but had low CUP1 copy number.2.2.9 Data AccessibilityTo facilitate data reuse, all genomic fastq files have been deposited in the NCBI-SRA database under theaccession code PRJNA261735. The remaining raw data and statistical analyses have been deposited in theDryad Digital Repository (doi:10.5061/dryad.5gp25).2.3 ResultsWe recovered a broad spectrum of genetic changes across 34 lines exposed to initially inhibitory levels ofcopper (Figure 2.1, Table 2.1). Most lines contained multiple mutations, in contrast to our previous resultsin nystatin (GERSTEIN et al. 2012), which is consistent with the longer waiting period before growth wasobserved (4-7 days with nystatin, 7-14 days with copper). All lines except for two (CBM2, one of the fivelines isolated on the first day, and CBM55, one of the eleven lines isolated on the last day) contained oneor more single base-pair mutations. In total, there were 57 unique base-pair changes, including four singlebase-pair deletions and one single base-pair insertion that also resulted in a basepair change. Beyond changesto single sites, there were several large-scale mutations. Twelve lines exhibited chromosomal aneuploidy(Figure 2.1B), and three lines (CBM6, CBM7, CBM17) appeared to have low mtDNA coverage, outside ofthe range of lines from our previous study with nystatin (Figure 2.1C). In addition, two changes involveddeletions within repetitive regions, one in CBM25 involving the tandem array of P-type ATPase sodiumpumps (ENA5, ENA2, and ENA1) and the second in CBM18 involving the flocculation gene FLO10 (seedetails in Table 2.1).The most frequent mutation across all lines was copy number alteration of the CUP1 locus. Based on insilico qPCR, CUP1 estimates were, on average, 2.17 times higher than estimates from the 35 lines obtainedin nystatin. CUP1 copy number was estimated to be above the entire range of nystatin lines for 24 of the 34CBM lines, while two lines (CBM16 and 26) both exhibited CUP1 levels lower than the range of nystatinlines (Figure 2.1C). CUP1 is a metallothionein protein that binds copper in S. cerevisiae. It is present as atandem duplication on chrVIII in the S288C reference strain, and amplification of this locus is known to bea common mutation that confers increased resistance to copper (ADAMO et al. 2012; FOGEL and WELCH1982; FOGEL et al. 1983). Disomy for chrVIII has also been shown to increase copper tolerance (FOGELand WELCH 1982). Indeed, including cases with chrVIII aneuploidy, 27 out of the 34 CBM lines haveincreased CUP1 copy number above the range of BMN lines (one of the exceptions, CBM24, is the leastcopper-tolerant of our CBM lines).When expression level of CUP1 was investigated by qPCR in a subset of lines and compared with the insilico qPCR estimates, it was found that the slope was positive and significant when forced through the point322.3. Results00.20.40.60.811.21.41.61.821 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16Bud neckBOI2* - CBM7BUD3 - CBM6BudBOI2* - CBM7VHT1* - CBM33NucleusGCR2 - CBM49FCP1 - CBM17YPL247C* - CBM20EndosomeDID4 - CBM46Vacuolar membraneMAM3 - CBM7, 14, 20VAM6 - CBM6VTC1* - CBM18, 22, 47VTC4 - CBM1, 5, 11,13, 17, 33$, 34$, 36$, 37$, 44$, Nuclear membraneESC1 - CBM17HMG2 - CBM49MLP1 - CBM25YPL247C* - CBM20Plasma membraneENA5 - CBM25PMA1* - CBM16, 20, 26, 29VHT1* -CBM33CytoplasmBOI2* - CBM7FAS1* - CBM1PFK1 - CBM21SBE22 - CBM33TRM7 - CBM7YPL247C* - CBM20Did not localizeFLO10 - CBM18MAL12 - CBM54Endoplasmic reticulumVTC1* - CBM18, 22,47GolgiSMI1 - CBM7MitochondrionFAS1* - CBM1PMA1* - CBM16, 20, 26, 29ChromosomeChromosome coverageCBM71722CBM717212226CBM21 CBM26CBM26CBM23111620212930B CAFES1 - CBM5145$, 460123456CBM24CBM54CBM51CBM22CBM21CBM55CBM53CBM4CBM2CBM49CBM6CBM30CBM25CBM37CBM29CBM36CBM3CBM26CBM7CBM5CBM1CBM14CBM34CBM33CBM16CBM18CBM44CBM45CBM47CBM20CBM13CBM46CBM11CBM17CUP1 coverageMitochondrial coverageFigure 2.1: Observed mutations in copper-adaptation lines. A. Genes mutated within the CBM lines (seeTable 2.1 for specific mutations) illustrated based on previous localization studies of the genes involved(references in Table 2.1). The colour of line names reflects the type of mutation; black for a nonsynonymousamino acid change, red for a premature stop codon, and blue for an indel or rearrangement (synonymouschanges and RNA genes not shown). Genes that localized to more than one location are listed multiplyand identified by (*). CBM lines 33, 34, 36, 37 and 45 contained the same VTC4 mutation, indicated by($). B. Chromosomal aneuploidy was prevalent, appearing in 11 of 34 lines. Coverage is determined by theaverage number of reads across the chromosome, compared to the reference strain. C. CUP1 copy number(green line) and mitochondrial coverage (orange line) for each CBM line. CUP1 copy number is measuredrelative to the average level observed in our parallel nystatin study, which showed negligible variation incopy number (range of the 35 BMN lines shown as dashed green lines). Mitochondrial depth of coverage inthe CBM lines was divided by ten and is presented relative to the average depth of coverage from mappednuclear DNA (the equivalent range for the 35 BMN lines shown as dashed orange lines). Lines are orderedaccording to increasing copper tolerance (Figure 2.2).332.3. Results(1,1), which assumes that both axes are scaled to the ancestor (even though, technically, the derived nystatinlines and not BY4741 were used as the control in the in silico qPCR assays; p = 0.02, Figure A.2A). Theslope was still positive, but not significant, otherwise (p = 0.27).2.3.1 Single base-pair changesOf the 57 unique single base-pair changes (Table 2.1), 15 were present in intergenic regions (two of whichwere indels), and one single base-pair deletion was present in the intron of COX1. The remaining 41 uniquesingle base-pair mutations were found within 29 different genes, whose products localize to many differentcellular structures (Figure 2.1A). Five of these mutations were synonymous changes, 24 were nonsynony-mous changes, 10 were premature stop codons, and two were frameshift mutations caused by single base-pairdeletions.Four sites were altered in the exact same way in multiple lines (a mutation at the intergenic site X.654261in three lines, mito.24277 within an intron of COX1 in seven lines, VII.480463 causing an amino acid changein PMA1 in two lines, and X.412080 causing a stop codon in VTC4 in six lines). As discussed previously(GERSTEIN et al. 2012), we cannot distinguish between repeated mutational hits and either a single ancestralmutation that amplified during the growth of the strain prior to being separated into lines (i.e., during growthof the ancestral colony, followed by overnight growth in 10mL YPD) or contamination during the samplingof lines on previous days. To be conservative, we consider the same mutation in multiple lines to be non-independent and count them as having arisen only once in the statistical analyses below.Four genes acquired multiple independent mutations, involving different positions in different strains.Fourteen lines acquired mutations in one of two subunits of the vacuolar transporter chaperone complex,VTC1 (3 unique mutations in 3 lines) or VTC4 (7 unique mutations in 12 lines); four lines acquired mutationsin the plasma membrane H+-ATPase PMA1 (3 unique mutations); and three lines acquired unique mutationsin MAM3, a protein required for normal mitochondrial morphology (ENTIAN et al. 1999).Given the 6607 ORFs within the S. cerevisiae genome (http://www.yeastgenome.org/cache/genomeSnapshot.html), the data are enriched for multiply hit genes. Specifically, there is a 99% chancethat the 41 genic mutations would either hit different genes (first line in equation 2.1) or would hit one genetwice but no more (second line in equation 2.1):0.99 =41Yi=16607 (i 1)6607+41Xj=20@j1Yi=16607 (i 1)66071A j  166070@ 41Yi=j+16607 (i 2)66071A , (2.1)assuming that ORFs are roughly equal in length. Thus, seeing even one gene bearing mutations at three ormore independent sites is highly unlikely, and we conclude that positive selection acted upon the mutationsin VTC1, VTC4, PMA1, and MAM3.Excluding the indels, the single base-pair mutations that occurred within exons generated a stop codon342.3. Resultsmuch more often than predicted by chance (10/39 = 25.6%, p = 0.00006, exact one-tailed binomial test withexpectation of 5.78% based on the mutational spectrum in yeast, see Materials and Methods). This resultremains marginally significant when we focus only on genes hit once and exclude the four multiply hit genes(4/25 = 16%, p = 0.053, expectation of 5.78%).On the other hand, the fraction of unique mutations that fall within an exon rather than a non-codingregion is not significantly greater than the expected fraction in S. cerevisiae (41/57 = 72.0%, p = 0.63, ex-pectation of 72.9% from ALEXANDER et al. 2010). Similarly, among the synonymous and non-synonymousmutations, non-synonymous changes did not occur more often than expected (including all changes: 24/29= 82.8%, p = 0.34; excluding multiply hit genes: 16/21 = 76.2%, p = 0.67; both exact one-tailed binomialtests with an expected fraction of 77.6%). Furthermore, the mean exchangeability score (an empirically-based measure of the change in protein function following a particular amino acid change, YAMPOLSKYand STOLTZFUS 2005) of our observed amino acid changes (0.294) was within one standard error of thegrand mean for mutations in yeast (0.31, calculated based on the mutational spectrum reported in LYNCHet al. 2008). These tests are likely conservative, however, because selection against deleterious amino acidchanges would have eliminated non-synonymous mutations from our dataset, making it difficult to detect anenrichment of amino acid changes due to positive selection.The set of genes whose protein products were altered is not enriched for either a specific GO termor a particular pathway (based on YeastMine analysis, BALAKRISHNAN et al. 2012), although a significantnumber of mutated genes localize to the plasma membrane (PMA1, ENA5, and VHT1), the nuclear membrane(ESC1, HMG2, MLP1, and YPL247C), and the vacuolar membrane (MAM3, VAM6, VTC1, and VTC4). Theset of genes is also enriched for three of the MIPS functional classification groups: vacuole or lysosome(VAM6, VTC1, and VTC4), cation transport (ENA5, PMA1, and VTC1), and protein synthesis (TRM7 andFES1) (identified using Funspec, ROBINSON et al. 2002).Of the characterized genes that bore mutations, 21 were available from the yeast knockout collection(GIAEVER et al. 2002). Relative to BY4741, 13 lines showed a significant increase in copper tolerance,and two showed a significant decrease in copper tolerance (Figure A.3). This assay supports the idea that anumber of the singly-hit genes might contain mutations that influenced copper tolerance.To identify potential regulatory changes caused by the 15 intergenic mutations we found, we assessedwhether predicted transcription factor (TF) binding sites were gained or lost using Cis-BP (WEIRAUCH et al.2014) (Table A.3). One of the positions (in CBM1) is not predicted to be at a TF binding site, while the re-maining 14 were split among changes that caused both gains and losses (five mutations), only gains (fourmutations), and only losses (five mutations). We identified a number of commonalities among the muta-tions, including two sets of transcription factor binding sites that were each lost together three times (one setinvolved members of the Forkhead family, FKH2 and HCM1; and a second set involved NHP6A, NHP6B,and PHO2), two that were gained together three times (GAT1 and GLN3, members of the GATA family),and some that were both gained and lost (particularly ORC2 and SUM1). Among TF binding site mutationsthat were within 500bp and 5’ of the start site of a gene, only one gene (RPP1A) was listed in the “Sac-charomyces Genome Database” (SGD) as having an effect on metal tolerance, although two others affectedvacuolar functioning (MUK1 and YIR007W) and one affected mitochondrial functioning (COX4). We did352.3. Resultsnot, however, directly measure the effects of the intergenic mutations.2.3.2 AneuploidiesChromosomal aneuploidy was common, appearing in one-third of all CBM lines (Figure 2.1B). All aneu-ploid lines had an extra copy of either chrII or chrVIII (one line had both). chrII aneuploidy was generallyfound by itself, only one of the eight lines with chrII aneuploidy carried additional aneuploid chromosomes(chrIII and chrVIII). In contrast, chrVIII aneuploidy never appeared in isolation. Three of the five cases ofchrVIII aneuploidy also contained an extra copy of chrXVI and one line contained additional copies of chrIand chrV (in addition to the aforementioned line containing chrII and chrIII).2.3.3 Mutagenic effects of copperWhile selection must underlie the repeated spread of mutations affecting the genes that were multiply hit,it is possible that copper exposure directly altered the rate and nature of mutations that arose during theexperiment. Indeed, exposure to high concentrations of copper is known to be mutagenic in experimentsthat directly expose DNA to copper (TKESHELASHVILI et al. 1991). There is no evidence, however, for anelevated base-pair mutation rate in our experiment. Focusing only on nucleotide changes (not indels), weobserved 52 unique single base-pair changes across the 34 lines isolated over the course of 7-14 days (average11.0 days until isolation, Table A.1). By comparison, in our previous study where the same ancestral strainwas exposed to nystatin, we observed 35 mutations among 35 lines isolated over the course of 4-7 days(average 4.7 days until isolation). Thus, if anything, slightly more mutations accumulated per line per day innystatin (0.21) than in copper (0.14), although the difference is not significant (p = 0.076, two-tailed exactbinomial test with n = 52 + 35 mutation events and a proportion expected in copper given by 0.693 giventhat there were 34 lines ⇥ 11.0 days in copper and 35 ⇥ 4.7 in nystatin). Furthermore, while previous invitro work indicates that copper should induce an excess of G:C! A:T mutations (TKESHELASHVILI et al.1991), the spectrum of single base-pair mutations observed within this study (8 A:T! G:C, 14 G:C! A:T,6 A:T! T:A, 12 G:C! T:A, 5 A:T! C:G, 7 G:C! C:G) is not significantly different from the mutationalspectra for yeast reported by LYNCH et al. (2008) (see their Table 1), either based on prior studies (2= 6.76,df = 5, p = 0.239) or based on their mutation-accumulation experiment (2= 1.48, df = 5, p = 0.915) .We did, however, observe many more aneuploid events with copper (affecting 12/34 lines) than withnystatin (affecting 1/35 lines, GERSTEIN et al. 2012), but this is only marginally significant if we accountfor the greater number of days until isolation (p = 0.11, two-tailed exact binomial test with n = 12 + 1aneuploid lines, where the proportion expected in copper is 0.693). Here, we have treated multiple aneu-ploid chromosomes within a line as a single event, in the absence of information about their independence;if they were independent, the excess of aneuploid events in the presence of copper would be very significant(p = 0.013, two-tailed exact binomial test with n = 19 + 1 aneuploid chromosomes). An enrichment ofaneuploid events in copper may well be due to selection for aneuploidy rather than an increased mutationrate, consistent with the frequent occurrence of additional copies of chrVIII, bearing CUP1. Nevertheless,362.3. Resultsprevious studies with mice have found copper to be mutagenic using a micronuclei assay that is sensitive toerrors in chromosome segregation during mitosis (PRÁ et al. 2008). We thus consider it plausible that thehigh frequency of aneuploidy observed in this study may have been directly due to copper exposure.2.3.4 Phenotypic assays of CBM linesCopper tolerance (measured as IC50 in deep-well boxes grown for 72 hours) was fairly similar across the34 copper-adapted CBM lines, ranging from 8.5mM – 11.2mM (Figure 2.2A). The date that mutations wereisolated does not correlate with copper tolerance (mutations are numbered based on the date of isolation;Table A.1). The number of copies of CUP1 inferred from in silico qPCR (Figure 2.1C, corrected to includechrVIII aneuploidy) does not directly correlate with copper tolerance (Figure 2.2B; r = 0.16, t32 = 0.92, p= 0.36), yet this is likely due to the confounding effects of the other genetic changes. For example, threeof the lines with the lowest CUP1 copy number carried mutations in PMA1 (excluding CBM24, which hadlow tolerance; Figure 2.2B). To tease apart the effects of these mutations, we both statistically analyzed thetolerance data collected for all lines and physically dissected the mutations via tetrad analysis in a subsampleof four CBM lines (see below). A linear model with the four multiply-hit genes as well asCUP1 copy numberand chrII aneuploidy as factors indicated that CUP1 copy number (adjusted to include chrVIII aneuploidy)as well as the presence of a mutation in MAM3, PMA1, VTC1, or VTC4 were all significant predictors ofcopper tolerance, while chrII aneuploidy was not (CUP1 coverage: t27 = 2.58, p = 0.016; VTC1: t27 = 3.73,p = 0.0009; PMA1: t27 = 2.99, p = 0.0060; MAM3: t27 = 3.11, p = 0.0044; VTC4: t27 = 6.66, p < 0.001;chrII: t27 = 1.67, p = 0.11).In addition to copper tolerance, we assayed maximum growth rates in copper, as well as YPD, YPG,and iron, using the Bioscreen C plate reader (Figure 2.3). Copper tended to be more inhibitory in the smallvolume plates used in the Bioscreen, so we reduced copper levels to 8mMCuSO4 (‘copper8’). Growth rate incopper8 (Figure 2.3A) was significantly correlated with copper tolerance (Figure 2.2A , r = 0.69, t32 = 5.40,p < 0.0001). All lines except CBM24 grew significantly faster in copper8 than did BY4741, the ancestor (p< 0.05, Table A.4). Copper tolerance did not correlate with growth in any of the three other environmentsexamined (Figure 2.3B-D; YPD: r = -0.31, t32 = -1.83, p = 0.077; YPG: r = -0.06, t32 = -0.34, p = 0.73; iron:r = -0.09, t31 = -0.52, p = 0.61). No lines exhibited significantly increased growth in the rich medium, YPD,while about half had significantly decreased growth (Table A.5). There was a negative, but not significantcorrelation between growth in YPD and growth in copper or copper tolerance (IC50). Three of the fourslowest growing lines in YPD carried multiple aneuploid chromosomes (CBM17, 22, 26), but otherwise theslow growing lines spanned a range of genotypes and CUP1 copy numbers. There was greater variation ingrowth rates observed in YPG and in iron, with growth in these two environments being strongly correlated(r = 0.58, t31 = 3.98, p = 0.0004, YPG statistical results in Table A.6, iron results in Table A.2). The fifteenCBM lines that grew significantly slower in YPG and iron included all lines with PMA1mutations (CBM16,20, 26, 29), chrXVI aneuploidy (CBM7, 17 and 22), and chrII aneuploidy (CBM2, 3, 11, 16, 20, 21, 29, 30),as well as all lines that lacked mtDNA (CBM16, 20, 29).372.3. ResultsA 6789101112BY4741CBM24CBM54CBM51CBM22CBM21CBM55CBM53CBM4CBM2CBM49CBM6CBM30CBM25CBM37CBM29CBM36CBM3CBM26CBM7CBM5CBM1CBM14CBM34CBM33CBM16CBM18CBM44CBM45CBM47CBM20CBM13CBM46CBM11CBM17Copper tolerance (IC 50 in mM)●●●●●●●● ●●●●●● ●●●● ●●●●● ●●●B 67891011120 1 2 3 4CUP1 coverage (vs. BY4741)●●●●+chr2MAM3PMA1VTC1VTC4Figure 2.2: Copper tolerance across 34 copper-adapted lines (‘CBM lines’). A. Lines are numbered based onthe date the mutant line was isolated following exposure to copper. The order of CBM lines in other graphsis based on the order of copper tolerance depicted here, which gives the IC50 after 72 hours of growth indeep-well boxes (bars represent 95% confidence intervals). Tolerance of BY4741 (the ancestor) is indicatedby the horizontal red line with its confidence interval indicated by dashed grey lines. B. Copper tolerance ofeach CBM line is generally high, regardless of the CUP1 copy number (x-axis, accounting for duplicationof chrVIII). The absence of a correlation between CUP1 level and copper tolerance is due to the existenceof additional mutations in the CBM lines, particularly in the four genes that were mutated independently(colors). Grey shading shows the range of CUP1 observed among the BMN lines (see Methods and Figure2.1C).382.3. Results●●●●●●●●●●●●● ●●●●●●●● ●●●●●●●●●●● ●● ●0.000.050.100.150.20A 8mM copper●●● ●●●●●●●●●●●●●● ●●●● ●●●●●● ●●● ● ●● ●●0.000.050.100.150.200.250.30B YPD (rich medium)● ● ● ●● ●● ● ●●●●●●●●●●●●● ● ● ● ●●● ● ● ●●● ●●●0.000.050.100.15C YPG (forced respiration)Maximum growth rate (/h)● ● ● ●●●● ● ● ●●●●● ●●●●●●● ● ● ● ●●●● ●●● ●●●0.000.050.100.150.200.250.30BY4741CBM24CBM54CBM51CBM22CBM21CBM55CBM53CBM4CBM2CBM49CBM6CBM30CBM25CBM37CBM29CBM36CBM3CBM26CBM7CBM5CBM1CBM14CBM34CBM33CBM16CBM18CBM44CBM45CBM47CBM20CBM13CBM46CBM11CBM17D 40mM ironFigure 2.3: CBM lines have variable growth rates under different environmental conditions. Maximumgrowth rate (±1 SE) was assayed within the Bioscreen C in four different environments, each on a singleday: A. YPD + 8mM CuSO4, B. YPD, a standard laboratory rich medium, C. YPG, a medium that forcesrespiration, and D. YPD + 40 mM ferric citrate. Closed circles are lines that are significantly different fromthe wildtype, BY4741 (red dashed line; see Supplementary Tables A.2, A.4-A.6 for statistical information).Vertical grey dashed lines are for ease of comparison among the panels.392.3. Results2.3.5 Tetrad dissections to isolate single mutationsTo examine the specific effect of commonmutations, we crossed CBM2 (chrII aneuploidy), CBM14 (MAM3),CBM25 (MLP1 and ENA5), and CBM34 (VTC4) to BY4739 and dissected the resulting tetrads. As indi-cated above each panel in Figure 2.4, the four mutations and chrII aneuploidy segregated according to a 2:2pattern (+/), but segregation of CUP1 copy number was more variable (shown by coloured circles). Thevariability in CUP1 inheritance may reflect noise in the assay (band densities on Southern blots, even thoughmeasured in triplicate, Figure A.2B), or it may indicate changes in CUP1 over the course of the tetrad lineconstruction. Indeed, previous work has shown copy number alterations were frequent (20%) during meiosisin lines heterozygous for CUP1 (WELCH et al. 1991).● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ●●●●●●●● ●● ●3:c(length(range) + 2)0.000.050.100.15●●+chrIICUP1ï + ï + ï + + ï ï + ï + + + ï ï+ïBY4739 CBM2 t1 t2 t3 t5A● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ●●●●●● ●●●●●●● ●●●●●●●3:c(length(range) + 2)0.000.050.100.15●●mam3CUP1ï + + ï ï ï + + ï + ï + ï + ï + + ï ï + + ï + ï+ïBY4739 CBM14 t1 t2 t8 t9 t10 t11B● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●●●●●●●●● ●●3:c(length(range) + 2)0.000.050.100.15●mlp1ena5CUP1ï + + ï + + ï ï + + ï ï + ï + ïï ï + + + ï ï + ï ï + + ï ï + ++ï +ïBY4739 CBM25 t1 t2 t6 t7C● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●●●●●●●● ●●●●●●● ●●●●●●●0.000.050.100.15●●vtc4CUP1+ ï + ï ï + ï + ï ï + + ï ï + + + ï ï + ï + ï ++ïBY4739 CBM34 t2 t3 t4 t6 t10 t11DMaximum growth rate in 9mM copper (/h)min CUP1max CUP1Figure 2.4: Maximum growth rates of tetrad lines in YPD+9mM CuSO4. Tetrads were derived from fourdifferent CBM lines: A. CBM2, B. CBM14, C. CBM25, and D. CBM34. For each line, maximum growthrate was assayed within the Bioscreen C, with bars representing ±1 SE across three days. The darkness ofthe circle represents the line’s relative number of copies of CUP1, as assayed by Southern blot. Presence(+) or absence () of a segregating mutation is also noted. The maximum growth rate is shown for all tetradlines, as well as for their two parents, BY4739, and the relevant CBM parent (red lines), except for the tetradsderived from CBM25 for which parental growth rate was not assayed (due to its initially being consideredCBM22, see Materials and Methods).Maximum growth rate in copper9 exhibited considerable variation among different spores from the sametetrad and rarely followed a strict 2:2 pattern (Figure 2.4), as expected when multiple mutations contribute to402.3. Resultscopper tolerance. The effect of each mutation on growth in copper is more easily seen in Figure 2.5, whichshows linear fits through all of the tetrad data for a given CBM line (using the mean maximum growth rateacross all replicates as a single data point for each spore). The interaction terms (seen as a difference in slope)were not significant, except for a marginally significant interaction between VTC4 and CUP1 (t20 = 2.044, p= 0.054, panel D), and thus interactions were excluded from the main statistical analyses. For CBM14 andCBM34, the presence of a mutation in MAM3 and VTC4 (respectively), as well as CUP1 copy number, hadsignificant positive effects on growth rate (CBM14, CUP1: t21 = 6.16, p < 0.0001, MAM3: t21 = 2.27, p= 0.034; CBM34, CUP1: t21 = 6.16, p < 0.0001, VTC4: t21 = 4.41, p = 0.0002). By contrast, only CUP1copy number had a significant effect on growth rate for CBM2 and CBM25 (CBM2, CUP1: t13 = 2.66, p =0.020, +chrII: t13 = 1.73, p = 0.11; CBM25, CUP1: t12 = 3.37, p = 0.0056, MLP1: t12 = 0.30, p = 0.77,ENA5: t12 = 0.018, p = 0.99).●●●●●●●●●●●●● ●●●0.000.050.100.152x105 3x105 4x105 5x105●●wt+chrIIA  CBM2● ●●●●●●●●●● ●●●●●●●●●●●●●2x105 4x105 6x105 8x105●●wtmam3B  CBM14●●●●●●●●●●●●●●●0.000.050.100.156x105 8x105 1x106 1.2x106 1.4x106●●wtmlp1C  CBM25●●●●●●●●● ●●●●●●●●●●●●●●●4x105 6x105 8x105 1x106●●wtvtc4D  CBM34Maximum growth rate (/h)CUP1 band intensityFigure 2.5: Impact of mutations on growth of tetrad lines in YPD+9mM CuSO4. Dots indicate maximumgrowth rates for tetrad lines carrying (red) or lacking (black) the mutation of interest: A. chrII aneuploidy inCBM2, B.MAM3 in CBM14, C.MLP1 in CBM25, and D. VTC4 in CBM34. For each line, linear model fitswere performed with maximum growth rate as the response variable and CUP1 copy number and the allelestatus of the other mutation(s) as the predictors. All fits are plotted including interaction terms betweenCUP1 copy number and the other gene of interest, whereMLP1 was used as the gene of interest for CBM25in panel C (no difference is seen when using ENA5).The growth rate in copper9 may reflect an overall growth impact caused by the mutations, rather than a412.4. Discussionspecific effect on growth in copper per se, interfering with our ability to detect improvement. To assess thispossibility, we measured growth in the rich medium, YPD, and reran the models performed above, usinggrowth rate in YPD as the response variable. The models for CBM14, CBM25 and CBM34 contained nosignificant terms (Table A.7). However, the model for CBM2 indicated that the presence of an extra copyof chrII contributed to a significant decrease in growth rate in YPD (Table A.7, Figure A.4). We thus reranthe copper9 linear growth models for CBM2, adjusting for growth rate in YPD by taking the difference.This model indicated that chrII aneuploidy indeed has a significant positive effect on growth in copper whencontrolling for its negative effect on growth in YPD (Table A.8).Together, these tetrad analyses indicate that CUP1 has a significant impact on growth in copper, as dothe mutations inMAM3 and VTC4. In addition, chrII aneuploidy has a significantly more positive impact ongrowth in copper than expected based on its negative effect on growth in YPD.Finally, from among all tetrads available for each line, we measured copper tolerance (IC50) for twospores that carried the mutation of interest (chrII aneuploidy,MAM3,MLP1, VTC4) yet exhibited low CUP1copy number (Figure A.1). All mutants were found to have a significantly higher copper tolerance thaneither the BY4741 ancestor or BY4739 parent (Figure A.5), despite their low CUP1 copy number. IC50 islikely to be a more sensitive assay of resistance to copper than maximum growth rate in a single copper level,suggesting that all of these mutations increase copper tolerance.2.3.6 Reexamining the petite mutationsAfter the above analyses were conducted, we reexamined the lines that displayed small colonies on YPDplates (Table A.1). To confirm that they were incapable of respiration, we assayed their growth on YPGplates. While 11 lines showed no growth, three lines (CBM9, 27, and 28) formed colonies, indicating thattheir mitochondria were still functional. These three lines were then whole-genome sequenced using pop-ulation samples (Table A.9). Line CBM9 carried a high-frequency SNP in PMA1, CBM27 carried a threebase pair deletion in PMA1, and CBM28 was aneuploid for chrIII, V, and VIII (a combination not seen inany other line). Levels of CUP1 were assayed as in Figure 2.1C and fell within the range of the BMN lines(CBM9: 0.69; CBM27: 0.83; CBM28: 1.25), except for CBM28 once we account for its extra copy ofchrVIII. Altogether, these lines provide additional confirmation of the role of PMA1 and chrVIII aneuploidyin the evolution of copper tolerance, although we infer that the mutational routes taken by these three lineswere particularly deleterious in the absence of copper, given their small colony size on YPD plates.2.4 DiscussionDepending on the different possible directions in which the environment may change, how rapidly can evo-lution happen and via how many different possible pathways? The advent of rapid sequencing technologyhas led to an increase in studies examining the repeatability of evolution at the level of the genotype, find-ing that the same genes often underlie parallel and convergent evolution in natural populations and during422.4. Discussionexperimental evolution studies (CONTE et al. 2012; MARTIN and ORGOGOZO 2013, see references within).Whether this repeatability reflects convergence over time to fitter genotypes or a limited scope of adaptivemutations along the way remains unknown. Also unknown is the relationship between the type of evolution-ary challenge and the likelihood of parallel genetic changes (STERN 2013). This study aimed to contributeto our understanding of how the type of environmental challenge influences the genomic target size of themutations selected during the very first steps of adaptation. We used the same mutation acquisition protocolas in our previous study on nystatin resistance (GERSTEIN et al. 2012) to obtain mutations in the presence ofan inhibitory level of copper. We predicted that a broader range of genetic solutions would underlie copperadaptation, in contrast to the nystatin study that identified a narrow genetic solution (all 35 lines had muta-tions within four genes in the same pathway). Previous studies have suggested that xenobiotic environments(such as antimicrobial drugs) select for repeated genetic solutions (MARTIN and ORGOGOZO 2013). By con-trast, copper is a very different kind of environmental stressor—it is essential for several different enzymaticprocesses in yeast (GRADEN and WINGE 1997) and therefore cannot be blocked entirely from entering thecell. Copper is, however, toxic at high concentrations (PEÑA et al. 1999), and thus its concentration in thecell must be held in a delicate balance. As predicted, we identified a large number of mutations among ourcopper adaptation lines, with the level of genetic parallelism highly dependent on the type of mutation underinvestigation.Increased copy number of the CUP1 locus through tandem duplication or aneuploidy of chrVIII wasby far the most common mutation, seen in 27 of the 34 copper adaptation lines. CUP1 exists as a tandemrepeat in the S288C genome, and adaptation under copper stress has previously been shown to select foramplification of this locus (ADAMO et al. 2012; FOGEL and WELCH 1982). Indeed, COVO et al. (2014)demonstrated that moving CUP1 onto other chromosomes can efficiently select for disomy under copperstress. The number of CUP1 copies varies among naturally-isolated wild and vineyard strains (between1 and 18 copies among 14 wild strains, ZHAO et al. 2014, and 4 and 18 copies among 15 Italian vineyardstrains, STROOBANTS et al. 2008). AsCUP1 is present in the ancestral genome as a tandem repeat, it is likelyprone to alterations in copy number as a consequence of unequal crossover, gene conversion, or single-strandannealing (ZHANG et al. 2013). Furthermore, CUP1 amplification seems to incur few pleiotropic costs, asseen by the lack of an observed effect of CUP1 copy number on growth rate in YPD among our tetrad lines(Table A.7).Chromosomal aneuploidy also repeatedly arose within our lines. Twelve of the 34 lines were aneuploid,and each of these 12 lines contained either chrVIII aneuploidy (five lines) or chrII aneuploidy (eight lines).Chromosomal aneuploidy seems to be a common route to adaptation for fungal species reproducing asexuallyin a diverse array of environmental stressors such as drug resistance (SELMECKI et al. 2009; SIONOV et al.2010), high temperature (YONA et al. 2012), and salt (DHAR et al. 2011). Aneuploidy is an intriguingbeneficial mutation, as it has the potential to affect many genes simultaneously, yet has a much higherreversion rate than other types of mutations. Whether such a high degree of aneuploidy would have beenseen had our strains evolved for longer remains an open question. It may be that aneuploidy serves primarilyas a stop-gap adaptation until other beneficial mutations appear in the genome with fewer costs (YONA et al.2012). In our experiment, chrVIII aneuploidy may have been positively selected for its effect on CUP1432.4. Discussioncopy number, as seen by FOGEL and WELCH (1982). Similarly, chrII aneuploidy had a more beneficialeffect in copper (Figure 2.5, Table A.8) than expected based on its low growth in YPD (Table A.7), perhapsbecause it amplified genes contributing to copper tolerance, such as SCO1 and SCO2 (non-adjacent geneduplicates on chrII), which function in the delivery of copper to cytochrome c oxidase in the mitochondrialinner membrane. ChrII aneuploidy was also repeatedly observed by COVO et al. (2014) when investigatingchromosomes gained in response to copper stress. Whether the other aneuploid chromosomes had an effecton copper tolerance remains unknown. Repeated attempts to sporulate CBM22 (aneuploid for chrVIII andchrXVI) and CBM26 (aneuploid for chrI, chrV, and chrVIII) failed to yield tetrads.The major genetic contributors to copper tolerance besides CUP1 were four genes involved in maintain-ing plasma membrane potential (PMA1), vacuolar transport (VTC1, VTC4), and mitochondrial morphology(MAM3). These genes each bore several independent protein-coding mutations in different lines, which ishighly unlikely in the absence of selection. Indeed, of the seven lines that did not exhibit CUP1 amplification(relative to the range of BMN lines), six involved mutations in these genes (three in PMA1, one in VTC1,two in VTC4), with the remaining low-CUP1 line exhibiting little copper tolerance (CBM24).The types of mutations observed in VTC4 and VTC1 point to selection for loss of function in these genes,with several mutations inducing stop codons or frame shifts (Table 2.1). Furthermore, deletion of VTC1and VTC4, as well as MAM3, significantly increases copper tolerance (Figure A.3). By contrast, PMA1codes for a plasma membrane H+-ATPase that regulates cytoplasmic pH, and yeast are inviable when thisgene is deleted. High levels of copper have previously been shown to have a deleterious effect on plasmamembrane organization (FERNANDES et al. 2000), while strongly stimulating plasma membrane ATPaseactivity (FERNANDES and SÁ-CORREIA 2001). It thus seems likely that the mutations identified in PMA1alter, rather than inactivate, this protein, suggesting a gain (or fine-tuning) of function. Consistent with thisview, none of the PMA1 mutations involved stop codons or frame shifts (neither the four listed in Table 2.1,nor the two additional PMA1 mutations among the lines initially categorized as petites, Table A.9).In addition, 25 genes bore a single mutation in a single line in our experiment (Table 2.1). These uniquemutations were no more likely to be nonsynonymous than expected based on the mutational spectrum andno more likely to occur within exons than expected. That said, several of the unique mutations causedstop codons (4/25), a marginally significant excess. Furthermore, 12/18 of the deletion lines tested for thissubset of genes were found to have significantly altered copper tolerance when compared to BY4741 (FigureA.3), suggesting that at least some of the uniquely hit genes may be playing a role in copper tolerance.Alternatively, many of the mutations in singly-hit genes may have been neutral but spread via hitchhiking, apervasive phenomenon in other batch culture experiments (LANG et al. 2013).Beyond genetic changes, epigenetic changes may have contributed to copper adaptation. This possibilitywas not formally investigated in our study. We can, however, conclude that epigenetic change was not theprimary cause of adaptation, given that plausible causative mutations, involving either CUP1 or the fourmultiply-hit genes, occurred in every line except one (Table 2.1, Figure 2.2B). The exception, CBM24,was the least copper tolerant line and remains a possible candidate for epigenetic adaptation, although ourgenomic analysis may not have found all rearrangements or changes in hard-to-align regions.One question of interest is why our copper-adapted lines carried so many more mutations, on average,442.4. Discussionthan the nystatin-adapted lines studied previously (GERSTEIN et al. 2012). In no case did we see a BMNline that carried two mutations thought to be adaptive; all lines carried one and only one mutation in an ERGgene. By contrast, 21 out of the 34 CBM lines carried more than one mutation for which we have evidence ofa beneficial effect in copper (including the four multiply-hit genes, chrII aneuploidy, and CUP1 levels abovethe BMN range when including chrVIII aneuploidy; 15 out of 34 lines if we exclude chrII aneuploidy). Thegreater contribution of multiple beneficial changes to copper adaptation cannot be explained by a larger poolof large-effect beneficial mutations or a higher mutation rate, because it took longer to observe growth incopper (7-14 days) than in nystatin (4-7 days). One possibility is that many of the adaptive mutations maynot have been beneficial enough on their own to generate detectable growth; instead, they may have alloweda line to persist for longer or to expand slightly in population size, facilitating the appearance of subsequentlarge-effect mutations. An alternative possibility is that positive epistasis among mutations more stronglyfavoured the spread of secondary mutations. Our tetrad analysis provides some evidence for this possibility.Figure 2.5 shows that the effect of VTC4 on growth rate rises with CUP1 copy number among tetrads ofCBM34, a marginally significant positive interaction (t20 = 2.044, p = 0.054). Similarly, the benefits of chrIIaneuploidy also appear mild at low CUP1 levels and rise with increasing CUP1 copy number among CBM2tetrads, although this interaction is not significant (t12 = 1.61, p = 0.13). In accordance with either of theseexplanations (mutations facilitating subsequent adaptation or positive epistasis), VTC4 mutations occurredmore often among the CBM lines with higher CUP1 copy numbers (Table 2.1, Figure 2.2B).This study provides an in depth analysis of how a eukaryotic organism, like yeast, takes its first fewevolutionary steps towards tolerating an inhibitory but essential element, copper. Our genome-wide analysisof 34 strains found that adaptation often involved a common step (especially amplification of CUP1), butthat routes less taken were also available. These alternate routes often involved chromosomal aneuploidy ofchrII or chrVIII and four genes with roles in a wide variety of cellular functions – vacuolar transporters, mito-chondrial morphology, and cytoplasmic pH regulation. Compared to our previous study of nystatin resistance(GERSTEIN et al. 2012), the variety of genes allowing growth in copper suggests that altered environmentswith more widespread effects on a cell may also provide a broader genetic basis for evolutionary recovery.Previous longer-term experimental evolution studies with microbes (e.g., CONRAD et al. 2009; HERRONand DOEBELI 2013; KRYAZHIMSKIY et al. 2014; LANG et al. 2013; MILLER et al. 2011; TENAILLON et al.2012; WONG et al. 2012) have found a high degree of parallel evolution at the gene level, and our resultssuggest that this can be the case even among the very first mutations selected. Whether adaptive mutationsare likely to recur depends, in theory, on the effect size of the mutations as well as their mutation rate. In ourcase, amplification of CUP1 is both a relatively large effect mutation and easily acquired, contributing to itshighly repeated nature, but it was also aided by the beneficial effects of many other less repeated mutations.In short, adaptation to copper is both more and less repeatable than adaptation to nystatin, with adaptation viaCUP1 representing the route most commonly taken, but with mutations affecting a variety of other cellularprocesses providing a diversity of less travelled paths toward copper adaptation.45Chapter 3Widespread Genetic IncompatibilitiesBetween First-Step Mutations DuringParallel Adaptation of Saccharomycescerevisiae to a Common Environment3.1 IntroductionThe number of different evolutionary pathways available to populations adapting to a new environmentdepends on the range and characteristics of possible genetic solutions. Even populations adapting to the sameenvironmental challenge can diverge genetically from each other if different mutations happen to establish.The long-term impact of this initial divergence depends on the fitness interactions between the availablealleles that underlie adaptation to a given environment (“epistasis”). Epistasis can run the gamut from allelesthat interact positively and augment each others’ fitness (“positive epistasis") to those that have negativeeffects on fitness in the presence of each other (“sign epistasis” WEINREICH et al. 2005) (Fig 3.1).3.1.1 Epistasis and its role in evolutionThe nature of epistasis is critical to broad-scale evolutionary phenomena. If all possible alleles have the sameeffect in all genetic backgrounds, we might expect populations that diverge initially to converge to a similargenotype and/or phenotype over time at the fitness optimum. In contrast, if some alleles are beneficial onlyin certain backgrounds, early genetic changes will limit future genetic options, and populations may divergegenotypically and phenotypically. Thus, the shape and ‘ruggedness’ of the fitness landscape is directlydetermined by the prevalence of sign epistasis (DE VISSER et al. 2011; POELWIJK et al. 2007, 2011).The type of epistasis can also shape the rate of adaptation. In the case of positive epistasis, when earlymutations increase the beneficial fitness effects of subsequent mutations, adaptive evolution can accelerateover time. In contrast, when epistasis is negative, i.e., when first-step mutations reduce or oppose the ad-vantage of subsequent mutations, evolution will decelerate. The deceleration of adaptation over time hasbeen previously found in a number of experimental evolution studies (CHOU et al. 2011; KHAN et al. 2011;KRYAZHIMSKIY et al. 2014; SCHENK et al. 2013).Even the formation of new species rests upon epistasis between alleles present in different nascentspecies. A major driver of postzygotic reproductive isolation between species is the build up of Bateson-463.1. IntroductionFigure 3.1: Types of epistatic relationships between mutations. (a) The type of epistasis is observed as thefitness of the single beneficial mutations (A and B) relative to the double mutant (AB). No epistasis occurswhen log fitness effects are additive, as shown here (growth rate, our primary fitness measure, is calculatedon a log scale). (b) Example plot showing the method used in this paper to illustrate epistatic relationships.The y-axis gives maximum growth rate over 24 hours. Point colours indicate strain genotype, where thedouble mutant is black, the ancestor is grey, and each single mutant has a unique colour. Lines are drawnbetween genotypes that are a single mutational step apart. Without epistasis, the lines form a parallelogram.Epistasis is observed as a double mutant with increased fitness (positive epistasis, higher hollow circle) ordecreased fitness (negative epistasis, lower hollow circle).Dobzhansky-Muller (BDM) genetic incompatibilities. These incompatibilities represent reciprocal sign epis-tasis, where alleles that work well together within a species perform poorly when combined with alleles fromthe other species in a hybrid individual, leading to hybrid inviability or sterility (COYNE and ORR 2004).Sign epistasis between hybrids and their more fit parental population can also contribute to speciation byreducing gene flow in one direction. With enough such asymmetric barriers acting in opposite directions,gene flow may cease entirely between populations.All models of speciation agree that sign epistasis, and particularly reciprocal sign epistasis, is importantfor speciation, but they differ on why species carry different alleles. Among the models of speciation by nat-ural selection, the classic explanation, proposed by DARWIN (1859), is that populations diverge into speciesbecause they experience different environments and so adapt in ways that often do not work well together.Because of the focus on environmental differences, this explanation has become known as “ecological spe-ciation” (SCHLUTER 2009). A contrasting hypothesis, known as “mutation-order speciation” (SCHLUTER2009), focuses on the chance order in which mutations arise and spread in different populations when fac-ing the same selective environment. Even if the mutational steps that have occurred in each population areindependently beneficial, combining mutations across populations need not be.3.1.2 Determinants of epistasisThe specifics of the selective environment(s) likely have a major influence on the nature of epistasis betweenbeneficial mutations. In environments where adaptation can occur via the elimination of a single biosyntheticpathway, complete loss-of-function mutations at one step in the pathway may lead mutations in downstream473.1. Introductiongenes to become irrelevant to fitness. Indeed, BATESON (1909) originally coined the term ‘epistasis’ in1909 to describe this type of interaction, in which the action of one gene was blocked by that of another,and this is primarily how molecular geneticists continue to define the word (AVERY and WASSERMAN1992). Considering instead partial loss-of-function mutations, genotypes combining multiple mutations maybe more fit than single mutations if flow through the biosynthetic pathway is reduced by each additionalmutation. In either case, we would expect double mutants to have equal or greater fitness than single mutantsif knocking out a pathway is beneficial (as long as there are no pleiotropic effects beyond the pathway), andconsequently sign epistasis and reproductive isolation should not arise.On the other hand, if an intermediate phenotype is optimal in a particular environment, mutations thatare beneficial on their own may overshoot the optimum when combined, causing a reduction in fitness. Inthis type of environment, theoretical work predicts that sign epistasis should be particularly frequent betweenindependently selected mutations that have relatively large effects on the phenotype (FRAÏSSE et al. 2016).There is also increasing evidence that epistasis is more often negative for mutations in functionally-related genes. In a large-scale screen for genetic interactions where mutations in most of the 6000 genes inthe yeast Saccharomyces cerevisiae were tested pairwise in 23 million double mutants (including mutationsin both non-essential and essential genes, although excluding ~1000 genes), COSTANZO et al. (2016) foundthat combinations of genes involved in the same biological process were enriched for negative interactions.This enrichment suggests, counter to intuition, that strongly negative fitness interactions, of the form that giverise to reproductive incompatibilities, may be more likely to accumulate between populations experiencingthe same selective environment compared to those experiencing different environments.3.1.3 Reproductive incompatibilities in nature and in the labTo date, few incompatibilities between or within species have been genetically characterized, although re-cent advances in genomic sequencing technology have greatly aided the discovery of the genetic basis ofspeciation. For natural populations, the majority of incompatible alleles (‘speciation genes’) that have beencharacterized are found between species adapted to different local environments, presumably representingcases of ecological selection (documented in NOSIL and SCHLUTER 2011 Tables S1 and S2). For example,the build-up of a suite of plant-specific traits has allowed one species of Drosophila to utilize a different,normally toxic, host plant (MATSUO et al. 2007), and selection on soils of different salinity has caused theaccumulation of QTL associated with salt tolerance in a hybrid species of Helianthus sunflowers beyondwhat is found in its parental species (LEXER et al. 2004). In other cases, genetic incompatibilities betweennatural populations have been identified where there is no clear connection to the external selective environ-ment, including BDMs caused by the reciprocal silencing of alternative duplicate gene copies (BIKARD et al.2009) or the differential accumulation of selfish genes and suppressors (see examples in MAHESHWARI andBARBASH 2011). The exact history of selection is unknown in natural populations, thus it is difficult to knowwhether these cases represent mutation-order or ecological selection. Natural populations of yeast also showenvironment-specific genetic incompatibility (including one characterized two-locus BDM HOU et al. 2015)though, as in other taxa, we have no knowledge of the evolutionary history that led to these interactions.Experimental evolution studies allow direct control over the form of environmental selection, and sign483.1. Introductionepistasis has been found in some studies that combined mutations from populations adapted to both differentand similar selective environments. DETTMAN et al. (2008) evolved different populations of Neurosporacrassa to high salinity and low temperature. When the evolved strains were mated, lineages adapted to dif-ferent environments exhibited reduced reproductive success relative to matings between lineages adaptedto the same environment, and this reduction was consistent with the action of BDM incompatibilities. Aparallel study that examined populations of S. cerevisiae evolved to high-salinity and low-glucose for 500generations found very similar results (DETTMAN et al. 2007). Follow-up work identified a BDM incom-patibility between an allele of PMA1 (a proton efflux pump) that arose under high salt adaptation and anallele of MKT1 (a global regulator of mRNAs encoding mitochondrial proteins) that evolved in low glucose(ANDERSON et al. 2010). This was the first reported BDM interaction among known genes isolated fromexperimentally evolved strains, to our knowledge.Sign epistasis has also been documented when combining mutations between experimentally-evolvedpopulations adapting to the same environment. KVITEK and SHERLOCK (2011) investigated populations ofasexually-propagated haploid S. cerevisiae evolved under glucose limitation in continuous culture for 448generations (KAO and SHERLOCK 2008). Mutations in two genes,MTH1 andHXT6/HXT7, appeared severaltimes in independent lineages during the experiment, but never together. These mutations were shown to beindividually beneficial, but they had lower competitive fitness when combined in a double mutant than eithersingle mutant or the ancestor, showing reciprocal sign epistasis (KVITEK and SHERLOCK 2011). Negativeepistasis was also prevalent among five additional strains constructed to bear two adaptive mutations thatarose in different lineages, with significant negative epistasis in four out of the five comparisons, includingone example of sign epistasis (KVITEK and SHERLOCK 2011). CHOU et al. (2014) similarly investigatedepistasis using an engineered strain ofMethylobacterium extorquens with a modified central metabolism thatwas dependent on a foreign pathway artificially introduced on a plasmid. These bacteria were evolved for900 generations under conditions that utilized this pathway. All adaptive mutations decreased expression ofthe introduced pathway. Combining mutations, the authors found that expression levels were well predictedby the independent effects of each mutation but that expression mapped nonlinearly onto fitness, leading tosign epistasis in many cases. Collectively, these experiments demonstrate that BDMs can arise rapidly inexperimental evolution studies, either when populations experience different or similar selective pressures,providing support for both ecological and mutation-order speciation.3.1.4 Investigation of epistasis between first-step mutationsWhat remains unknown from long-term experiments evolved under the same selective pressure is how fre-quently early adaptive mutations could contribute to reproductive isolation. This raises the question ofwhether mutation-order speciation occurs because of incompatibilities among mutations that would be bene-ficial in either population or because the fixation of different initial mutations alters the subsequent selectiveenvironment experienced in different populations (i.e. divergent selection due to differences in genetic back-ground).We investigate, for the first time, fitness interactions among all pairwise combinations of genes bear-ing first-step adaptive mutations to a common selective environment. Specifically, we measured epistasis493.1. Introductionbetween beneficial mutations acquired in the yeast Saccharomyces cerevisiae grown in the presence of thefungicide nystatin (GERSTEIN et al. 2012). Briefly, GERSTEIN et al. (2012) isolated 35 first-step mutationsin 4 mM nystatin, performed genome-wide sequencing, and found that all strains carried a single mutationin one of four genes in the ergosterol biosynthesis pathway (Fig 3.2; genomic analysis revealed either no oronly one other mutation present in the strains used herein, details below). We focused on one mutation ineach gene and investigated the fitnesses of all six pairwise double mutants between these four mutations.For two of these genes (ERG6 (SGD ID: S000004467) and ERG3 (SGD ID: S000004046)), many of themutations found by GERSTEIN et al. (2012) were consistent with a complete loss of function (e.g., early stopcodons, similar sterol phenotype to the whole gene knockout). The mutations occurring in the most upstream(ERG7 (SGD ID: S000001114)) and downstream (ERG5 (SGD ID: S000004617)) genes in the pathway,however, were not (GERSTEIN et al. 2012). The erg7 mutation is a nonsynonymous change close to the endof the gene, and deletion of ERG7 is inviable. The erg5 mutation is an in-frame deletion and is unlikely tobe a null mutation because the full gene deletion is respiratory deficient (MERZ and WESTERMANN 2009),which is not observed for this mutant (BMN35 in GERSTEIN et al. 2012). Thus, we also assessed whetherupstream mutations in the biosynthetic pathway generally mask the effects of downstream mutations or ifmasking is limited to complete loss-of-function mutations.Figure 3.2: An abbreviated version of the ergosterol biosynthesis pathway. For each gene used in this study,we highlight its position in the ergosterol pathway, with gene names coloured according to the scheme usedin subsequent figures. Pathway adapted from LEES et al. (1995).Overall, we found that strong negative epistasis, of the type that causes some degree of reproductiveisolation, between strains fixed for different mutations was surprisingly common among these first-step mu-tations. Indeed, the interactions were so negative that they reversed the direction of effect in over half ofthe double mutants, causing beneficial mutations to become deleterious when in combination and doublemutants to be less fit than at least one of the two single mutants (sign epistasis) (Fig 3.1). Furthermore,in one-third of the comparisons, the double mutants were less fit than both single mutants (reciprocal signepistasis). We assayed mutational effects in both haploid and diploid backgrounds, finding similar results re-gardless of ploidy, indicating that these epistatic relationships are likely to hold across stages of the yeast lifecycle. Epistatic relationships for fitness were not well predicted by sterol profiles or pathway position of themutants, however, suggesting that selection does not simply act via flux through the pathway to ergosterol.Finally, we investigated epistasis in different concentrations of nystatin to determine how epistatic rela-tionships, and therefore reproductive isolation, might change under different levels of environmental stress.Previous work with antibiotic resistance in bacteria has shown that the shape of fitness landscapes can be503.2. Materials and Methodsstrongly dependent on antibiotic concentrations (MIRA et al. 2015). Interestingly, we found that the nega-tive interactions observed between beneficial mutations at lower concentrations of nystatin reversed sign andbecame increasingly positive at higher concentrations of nystatin. Indeed, only the double mutants exhibitedsubstantial growth in the higher concentrations of nystatin tested. Thus, while combining single-step muta-tions generally reduced fitness in the historical nystatin environment, these same combinations were morelikely than the individual mutations to allow colonization of even harsher environments.3.2 Materials and Methods3.2.1 Strain constructionWe assayed all pairwise interactions in both haploids and diploids between four beneficial mutations acquiredin the fungicide nystatin, one in each of ERG3, ERG5, ERG6 and ERG7 (Table 3.1). Each mutation wasinitially isolated in the BY4741 haploid background (MATa his31 leu20 met150 ura30, derived fromS288C) and struck down to a single colony to remove standing variation. Mutations were detected by wholegenome sequencing on an Illumina HighSeq 2000, followed by alignments to the S288C reference strain(GERSTEIN et al. 2012); few other mutations were detected besides those in the ERG genes. For the strainsused here, only the strain containing the mutation in ERG7 also carried a secondary mutation (inDSC2 (SGDID: S000005434)), whose presence or absence did not substantially alter the presented results (see detailsin Section B.1.2). For a complete description of the isolation of these initial strains, see GERSTEIN et al.(2012). All possible haploid and diploid genotypes for each pair of ERG genes were created via mating andsporulation. A brief overview of strain construction will be given here but for a detailed description, seeSection B.1.1.Table 3.1: Beneficial mutations in the strains used for the study of epistasis in the presence of nystatin(GERSTEIN et al. 2012).Strain Gene Genome Position Position in Mutation Amino Acid Change(Chr.Bp) Gene (nt)BMN1 ERG7 VIII.241194 2097 C->G Phe699LeuDSC2 XV.193885 916 G->A Asp306AsnaBMN9 ERG6 XIII.252612 379 G->C Gly127ArgBMN32 ERG3 XII.254758 898 G->C Gly300ArgBMN35 ERG5 XIII.302174 - 302233 253 - 312 60-bp deletiona Not known to affect fitness. Encodes a multi-transmembrane subunit of the DSC ubiquitin ligase complex(RYAN et al. 2012; TONG et al. 2014). Null mutant has decreased competitive fitness (BRESLOW et al. 2008)and decreased resistance to glycolaldehyde (JAYAKODY et al. 2011).To create singly heterozygous strains, each original single mutant strain was mated to BY4739 (MAT↵leu20 lys20 ura30) (Open Biosystems), which is isogenic with BY4741 except for the auxotrophies.MAT↵ single mutant strains were isolated by sporulation of the heterozygous diploids followed by dissection513.2. Materials and Methodsand testing of the resulting tetrads. Throughout strain construction, histidine and lysine auxotrophies wereconsistently kept with the same mating types so that all haploid strains were either MATa his31 or MAT↵lys20. Plates lacking methionine did not efficiently select against the met150 mutation carried by theoriginal single mutant strains, suggesting a weak effect of this mutation, and the methionine auxotrophy wasnot tracked.The MAT↵ single mutant strains were then mated to the original MATa single mutant strains to createstrains that were either homozygous for one mutation or heterozygous for two mutations. The haploiddouble mutant strains were created through sporulation and dissection of the doubly heterozygous strains.All haploid double mutant strains were confirmed by Sanger sequencing.We failed to obtain the MATa erg5 erg6 double mutant haploid strain through crossing and sporulationbecause the two genes are linked (within 48 kb and flanking the centromere of chr XIII). For this strain,a transformation was performed by electroporation using a protocol based on CREGG (2007) to insert themutation within ERG6 into the MATa erg5 genetic background; this insertion was then checked by Sangersequencing.Strains with one heterozygous and one homozygous mutant locus as well as double homozygous mutantstrains were created by mating theMATa single mutant and double mutant strains to theMAT↵ double mutantstrains.A representative of the diploid ancestral strain was created by mating BY4741 and BY4739.3.2.2 Growth rate assaysWe conducted a set of growth rate (fitness) assays under nystatin stress and in rich medium (YPD). Theexperimental design sought to ensure that data was gathered for each combination of wildtype and mutantstrains across batches performed on different days. Specifically, within a batch, for a given pair of mutationsin haploids and for each mating type, each ancestral strain and each single mutant was assayed twice, whileeach double mutant was assayed four times (the double mutant was assayed more often because it wasthe only genotype unique to that pair of mutations). For each pair of mutations in diploids, all possiblecombinations of the two genes in both heterozygous and homozygous forms (including the non-mutant)were present twice within a batch.We measured growth in YPD and YPD + 2 mM nystatin (‘nystatin2’) using the Bioscreen C Microbio-logical Workstation (Thermo Labsystems), which measures OD in 100-well honeycomb plates. Nystatin2was used to assay fitness because previous studies with these mutants found that 2 mM nystatin inhibitsthe growth of the ancestral strains while also allowing the growth of all mutant strains (GERSTEIN 2013).OD was measured automatically using the wideband filter at 30 minute intervals for 24 hours from culturesgrowing at 30°C with maximum continuous shaking. Longer assays were avoided because mutations andloss of heterozygosity events begin to accumulate (GERSTEIN et al. 2014). The maximum growth rate over24 hours was determined by the spline with the highest slope from a loess fit through natural log transformedOD data, using a custom R script written by Richard Fitzjohn (R CORE TEAM 2015) (see ONO et al. 2016for code).For complete details on how strains were initially grown from frozen and standardized (“pre-assays”)523.2. Materials and Methodsbefore measuring growth (“assays”), see Section B.1.3. Briefly, each yeast replicate was grown from frozenin YPD + 0.5 mM nystatin in 100-well honeycomb plates for 72 hours in the pre-assays, unless very poorgrowth of the strain required otherwise, and OD was then determined. YPD + 0.5 mM nystatin was used tohelp prevent reversion of strains with severe growth defects in YPD and was not found to affect subsequentmeasures of growth, compared to a pre-assay in YPD (the first pre-assay was conducted in YPD, see detailsin Section B.1.3). For the main assays, honeycomb plate wells were filled with 148.5 mL of YPD or nystatin2.The yeast were then transferred from the pre-assay plates into one well each of YPD and of nystatin2, withthe volume transferred determined by the maximum pre-assay OD reading (the minimum volume transferredwas 1.5 mL while the maximum was 7.5 mL). Note that these transfers decreased the concentration of nystatinin the individual wells, but never by more than 0.1 mM. Strains were randomized within plates using the samemap for the pre-assays and assays in a given batch.There were not equivalent numbers of replicates for all strains after omitting some data due to lowgrowth (if the volume to be transferred to the assay plate exceeded 7.5 mL), lack of growth, mechanical error,or because some strains had to be re-run (for details, see Table B.2). Nevertheless, at least two replicatesper day on at least two days were measured for all strains in each medium (with the exception of erg5/erg5erg6/erg6 for which 14 replicates were all run on a single day, Table B.2; for exact numbers and daysthe replicates were run on, see ONO et al. 2016). Although the different numbers of replicates led somecrosses to have less power than others, the cross with the least amount of data (erg6 by erg7) was also theone where the double mutant was particularly unfit, which contributed to the difficulties in assaying fitnessbut also meant that epistasis was readily detected. In all cases, data for each double mutant was collectedsimultaneously with data on the ancestor and single mutants, allowing day effects to be factored out in theanalysis.3.2.3 Tolerance across a range of nystatinGrowth at different concentrations of nystatin was assessed following similar procedures to the growth rateassays. To prepare the strains for tolerance assays, pre-assays were again conducted to standardize initialcell concentrations. Stocks were first grown from frozen in four 96-well plates filled with 198 mL of YPD +0.5 mM nystatin and inoculated with 2 mL of frozen culture. Strains were distributed among the four platesso that there was one replicate of the entire balanced design per plate, randomized within plate. In order tofit all strains on a single plate, some strains were excluded (MATa erg5 erg6 and MATa erg3 erg5). Thesestrains were chosen because initial assays indicated that these double mutants most closely resembled thestronger (non-erg5) single mutant. The plates were covered with aluminum lids and incubated at 30°C withcontinuous shaking at 200 rpm in a container with wet paper towels to minimize evaporative water loss. Priorto removal of the aluminum lid, plates were always spun for 1 min at 3700 rpm to ensure that all liquid wascollected at the bottom.After 72 hours, all wells were manually mixed and ODwas measured on a BioTek plate reader at 630 nm.The well with the minimum OD value among the four pre-assay plates was identified and used to calculatethe amount of YPD to add to each pre-assay well to standardize cell density across cultures. Wells containingonly medium, those containing erg6/erg6 erg7/erg7 (see below) and one well that appeared not to have been533.2. Materials and Methodsinoculated were excluded from standardization. 2 mL from each well was used to inoculate the assay plates.Assay plates were prepared with 198 mL of YPD + 0, 1, 2, 4, 8, 16, 32, 64, 128, and 256 mM nystatin, withfour plates per concentration. The assay plates were covered with aluminum lids and incubated at 30°C incontainers with wet paper towels, shaking at 150 rpm.Exceptions to the pre-assay protocol had to be made for strains with slower growth. 10 ml of 0.5 mMnystatin was inoculated with 15 mL of erg6/erg6 erg7/erg7 from frozen two days before all other strainswere inoculated, allowing additional growth time for this unfit strain. On the day that all other strains wereinoculated from frozen, the erg6/erg6 erg7/erg7 culture was concentrated into ⇠900 mL (although growthwas not observable), and 200 mL of this culture was used to replace the medium from the appropriate wellsin the pre-assay plates. In addition, erg6 erg7 (both MATa and MAT↵) and erg6/erg6 were inoculated with2.67 mL of frozen culture (as opposed to the 2 mL used for all other strains) to compensate for their lowergrowth rate from frozen.Twenty-four hours after inoculation, the aluminum lids were removed, wells were manually mixed, andthe OD of each assay plate was read on a BioTek plate reader at 630 nm. Some wells had lost volume dueto cracks that had developed in the plates, and these wells were omitted from analysis. Prior to analysis, theOD of the medium itself was subtracted from the final OD measurements.3.2.4 Sterol AssayTo determine whether the sterol profiles of the single mutants, along with their position within the ergosterolpathway, predict the sterol profiles of the double mutants and whether differences in sterol profiles predictdifferences in fitness, a spectrophotometry-based assay was used to compare the sterol profiles of the ances-tral, mutant and double mutantMATa strains. Sterols were extracted using the alcoholic potassium hydroxidemethod (ARTHINGTON-SKAGGS et al. 1999), as previously performed on the single mutation strains (GER-STEIN et al. 2012). MATa strains were struck from frozen onto YPD plates and grown for 65 hours. Threecolonies for each strain were inoculated into two separate tubes filled with 10 mL of YPD (total of 20 mLper replicate) and incubated at 30°C on a rotor for 48 hours.After growth, cells were harvested by centrifugation at 2700 rpm for 5 minutes, combining culture fromthe two tubes by performing two successive spins. The pellets were washed twice with sterile distilled water.1.2 mL of 25% alcoholic potassium hydroxide was added to each pellet, and the tubes were vortexed for1 minute. The tubes were then incubated in an 80°C water bath for 1 hour. After cooling the samples toroom temperature, 0.4 mL of sterile distilled water and 1.2 mL of n-heptane were added to each sample,and the tubes were vortexed for 3 minutes. Samples were collected by taking 220 mL of the heptane layerand adding it to 880 mL of 95% ethanol in a 1.5 mL tube. These tubes were stored at -20°C for two daysbefore reading the absorbance every 3 nm between 200 and 300 nm in a quartz microcuvette using a ThermoBioMate 3 spectrophotometer. Due to a posteriori observations that different heptane/ethanol mixtures ledto different peak heights near 220 nm, we chose to use one replicate of the erg6 erg7 strain that showed noevidence of growth (suggesting an inoculation failure), but was otherwise identically treated, as a control forstandardization. As a result, only two replicates of erg6 erg7 are presented.543.2. Materials and Methods3.2.5 Outlier detection and removalOutliers in microbial fitness assays often represent either contamination by a different strain or evolutionover the course of the fitness assay. In order to prevent these events from having undue influence on ouranalyses, we detected outliers for maximum growth rate after omitting some wells due to lack of growth andmechanical error (see details in Section B.1.3). For outlier detection, we first normalized for plate withineach day. We did so by finding the global mean maximal growth rate for all ancestral strains over all daysand calculating the difference between this and the mean of all ancestral strains on a given plate, yielding aplate correction value. This correction value was added to each strain from the corresponding plate. Outlierswere detected by performing a two-sided Grubbs test, allowing us to detect a maximum of one outlier perstrain and medium, using the R package outliers and the method grubbs.test (KOMSTA 2011; R CORE TEAM2015). A total of eight replicates in nystatin2 and six replicates in YPD were marked as outliers and removedfrom all presented statistical and graphical analyses.All qualitative relationships between strains and the main statistical conclusions were insensitive to theexclusion or inclusion of the identified outliers, with two main exceptions for the haploids in nystatin2 (seeFig B.6 and Fig B.7 for versions of Fig 3.3 and Fig 3.4 that include all outliers). These exceptions are notedin the Results and described in detail in Section B.1.4.3.2.6 Statistical analysesEpistasis for maximum growth rate was assessed with mixed-effects models run on either all haploid or alldiploid strains together, including the genotype at each gene, their pairwise interactions, and mating type(for the haploids) as fixed effects and plate within day as a random effect, fit using restricted maximumlikelihood with the lmer function from the lme4 package in R (BATES et al. 2015; R CORE TEAM 2015).For diploids, the models were first run using only strains that were homozygous (either mutant or ancestral)for comparison to the haploid data. Significance of interaction terms (and mating type) was determined byperforming an ANOVA between the full model and a model dropping that term using the anova function inR and fitting models using maximum likelihood.To determine the type of epistasis present for each pair of genes, the package lsmeans (LENTH 2016) wasused to both determine the least-squares mean for each strain in the model and to make comparisons betweenstrains using the contrast function. The type of epistasis was determined by comparing the double mutantto each single mutant and each single mutant to the ancestor, and only these planned comparisons wereperformed. The P-value was adjusted for the number of tests performed using the multivariate t distribution(mvt method) in lsmeans. To be conservative, we based our categorization of epistasis solely on statisticallysignificant differences. For example, if the double mutant had a lower growth rate than both single mutantsbut this difference was only significant in one of the two cases, it was considered an example of sign epistasis(significantly lower than one single mutant but not the other) rather than reciprocal sign epistasis.A similar procedure was then undertaken including heterozygous diploid strains. A model was run usingthe lmer function including all diploid strains together, with plate within day as a random effect. Least-squares means were determined for all diploid genotypes from this model, and comparisons were performedbetween each diploid genotype and all other diploid genotypes that were one mutational step away. The553.3. Resultsdouble heterozygous strains were compared to all other strains for that pair of genes because the potentialprogeny of the double heterozygote includes all possible genotypes and these comparisons are therefore ofbiological interest.For the tolerance assay assessed across a range of concentrations of nystatin, we performed Welch’s t-tests of OD after 24 hours between each double mutant and its single mutant parents (day effects were notestimated as all measurements were gathered on the same day). Because we were focused on the changingnature of epistasis, rather than any particular pairwise comparison, a correction for multiple comparisonswas not performed.Data and analyses deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.vs370 ONO et al. (2016).3.3 Results3.3.1 Epistasis of haploids in nystatinWe characterized the epistatic interactions between pairs of mutations that act in the ergosterol biosynthesispathway and individually confer increased fitness when exposed to the antifungal drug nystatin. Maximumgrowth rate of ancestral, single mutant, and double mutant genotypes was characterized in haploid strainsof both mating types in YPD + 2 mM nystatin (‘nystatin2’). Outlier data points were detected statisticallyand removed from further analyses, although we note where inclusion of outliers would have affected theresults (for further details, see Section 3.2.5). The effect of mating type (and its associated auxotrophy) wasnot significant (P = 0.19), and the data for the two haploid mating types will be considered together, exceptwhere noted (see ONO et al. 2016 for additional statistical methods and results).Using a mixed-effects model, all main effects of individual mutations were positive, confirming that themutations improved growth in nystatin (Table 3.2). Double mutants were never significantly more fit thanthe best of the single mutants (top right panels in Fig 3.3), and all pairwise interactions exhibited significantnegative (antagonistic) epistasis (Table 3.2). To assess epistasis, least-square means of maximum growthrates were inferred from the model and compared between double and single mutants and between singlemutants and ancestral strains, correcting for multiple comparisons. The double mutant was significantly lessfit than the fittest single mutant in four cases (“sign epistasis”: erg3 erg5, erg3 erg6, erg3 erg7 and erg6erg7) and significantly less fit than both single mutants in two cases (“reciprocal sign epistasis”: erg3 erg6and erg6 erg7, Table 3.2, Fig 3.3). The results are similar when fitness is measured by optical density after 24hours of growth instead of maximum growth rate over 24 hours (Fig B.1). The strong negative interactionsindicate that these alleles, when combined, confer genetic incompatibilities between the strains.3.3.2 Comparison of epistasis between haploids and diploidsWe characterized epistatic interactions of maximum growth rate for the homozygous diploid strains in nys-tatin2 and compared them to the haploid results to determine whether the interactions were ploidy-dependent.As in haploids, single mutations generally improved the growth of diploid homozygotes in nystatin2, al-though the erg5 mutation did not do so significantly in a pairwise comparison with the ancestral strain563.3. ResultsFigure 3.3: Maximum growth rate of haploid strains in nystatin2 (above diagonal) and YPD (below diago-nal). Points are the fitted least-squares means of the maximum growth rates, determined in the mixed-effectsmodel. ⇥’s denote the additive fitness null expectation for the double mutant, i.e., with no epistasis. Eachsingle mutant is coloured differently, the double mutant is black, and the ancestor is grey. Vertical barsrepresent 95% confidence intervals of the fitted least-squares mean. Solid lines indicate significant contrastsbetween the fitted means, while dotted lines are non-significant. Combinations showing significant sign (S)and reciprocal sign (RS) epistasis are indicated by the presence of the abbreviation at the top of the panel. Innystatin2, the comparison between erg3 erg5 and erg3 is not significant when outliers are included, and theerg3 erg6 vs. erg6 comparison is only marginally significant (P = 0.083). In YPD, comparisons erg3 erg6vs. erg6 and erg6 erg7 vs. erg7 are not significant when outliers are included. All underlying raw data andanalyses can be found in ONO et al. (2016).573.3. ResultsTable 3.2: Results from a mixed-effects model run on all genes using the haploid maximum growth ratedata in nystatin2. Coefficients of main effects are the differences in mean maximum growth rate betweenthe single mutant strains and the ancestral strain (difference betweenMAT↵ andMATa in the case of matingtype). Coefficients of interaction terms are the differences in mean maximum growth rate between the doublemutant strains and the sum of the two single mutant coefficients added to the ancestral value. P-values are theresult of an ANOVA between the full model and one lacking that term; significant P-values are in bold. Thelast three columns refer to the type of epistasis present (Fig 3.1). “Epistasis” indicates a significant departurefrom an additive model of growth rates, which can be either negative or positive. “Sign” and “Reciprocalsign” refer to cases where the double mutant grows significantly less well than one or both single mutants,respectively.Term Coefficient SE P Epistasis Sign Reciprocal signmating type -0.0034 0.0026 0.19erg3 0.18 0.0057erg5 0.030 0.0049erg6 0.15 0.0049erg7 0.10 0.0049erg3*erg5 -0.054 0.0090 3.1 ⇥ 109 negative aerg3*erg6 -0.20 0.0090 < 1015 negative aerg3*erg7 -0.18 0.0090 < 1015 negativeerg5*erg6 -0.031 0.0076 4.6 ⇥ 105 negativeerg5*erg7 -0.046 0.0078 5.1 ⇥ 109 negativeerg6*erg7 -0.18 0.0083 < 1015 negativeaNot significant when outliers are included.(Fig 3.4). Qualitatively, epistatic interactions were also similar to the haploids (Table 3.3, Fig 3.4), whetherfitness was measured by maximum growth rate or optical density after 24 hours of growth (Fig B.2).When we categorized the type of epistasis statistically for maximum growth rate, most interactions wereof the same type (sign epistasis: erg3 erg5; reciprocal sign epistasis: erg3 erg6 and erg6 erg7; negativeepistasis: erg5 erg7). There were, however, several quantitative differences. The erg6 erg7 double mutantwas so unfit in diploids that we were often not able to standardize it properly in the growth assays (lowgrowth, as measured by optical density, was observed in all concentrations of nystatin tested, Fig B.3).Furthermore, in two cases, epistasis was qualitatively similar, but the differences were no longer statisticallysignificant (sign epistasis: erg3 erg7; negative epistasis: erg5 erg6).To visualize the full diploid fitness landscape, we repeated the analysis including all heterozygous strains(open symbols in Fig 3.4, pairwise comparisons in Fig B.4). Low F1 hybrid fitness was typical; double het-erozygous strains (open diamonds) were uniformly low in fitness when compared to the homozygous singlemutants (not significantly so when compared with the weak erg5/erg5 mutant). Mutations were generallypartially to fully recessive and did not have a large effect on fitness when comparing heterozygotes to wild-type at a gene, either when the other gene was wildtype (open triangles) or homozygous mutant (opencircles).583.3. ResultsFigure 3.4: Maximum growth rate of diploid strains in nystatin2 (above diagonal) and YPD (below diagonal).Points are the fitted least-squares means of the maximum growth rates, with closed circles determined in themixed-effects model including only homozygous strains and open symbols from the model that includesheterozygous strains (open diamonds: double heterozygotes; open triangles: single heterozygotes that arewildtype at the other gene; open circles: single heterozygotes that are homozygous mutants at the othergene). Points and bars are otherwise as in Fig 3.3. All symbols are coloured intermediately according togenotype and arrayed along the x-axis so as to lie between the two strains that are genotypically most similarto it. Solid lines indicate significant comparisons in tests run including only homozygous strains while dottedlines are non-significant comparisons. See Fig B.4 for statistical comparisons including heterozygous strainsand Fig 3.3 for further graphical details. In YPD, the homozygous comparison erg3 erg5 vs. erg3 is notsignificant when outliers are included. Note that the point for erg5/ERG5 erg6/erg6 was removed because itwas later found to have lost heterozygosity at ERG5. All underlying raw data and analyses can be found inONO et al. (2016).593.3. ResultsTable 3.3: Results from a mixed-effects model run on all genes using the homozygous diploid maximumgrowth rate data in nystatin2. For statistical and column details, see Table 3.2.Term Coefficient SE P Epistasis Sign Reciprocal signerg3 0.18 0.0065erg5 0.0028 0.0058erg6 0.16 0.0060erg7 0.088 0.0057erg3*erg5 -0.043 0.012 0.00037 negativeerg3*erg6 -0.22 0.012 < 1015 negativeerg3*erg7 -0.12 0.012 < 1015 negativeerg5*erg6 -0.015 0.012 0.19erg5*erg7 -0.025 0.010 0.015 negativeerg6*erg7 -0.26 0.014 < 1015 negative3.3.3 Epistasis for growth in YPDTo determine the extent to which epistasis reflected gross fitness defects not specific to nystatin resistance,we repeated the analysis on maximum growth rate in YPD, a rich growth medium. As in nystatin2, matingtype (and its associated auxotrophy) had no significant effect (P = 0.98), and results were averaged overmating types.The single mutations were generally deleterious in YPD (note the negative coefficients for the individualmutations, Table 3.4 and Table 3.5), consistent with previous characterization of these mutations (GERSTEINet al. 2012). The exception is the haploid erg5 mutant, which is not significantly less fit than the ancestorin a pairwise comparison of maximum growth rates (bottom left panels in Fig 3.3, Fig 3.4). As observed innystatin2, the double mutant often had lower fitness than the single mutants in YPD, although the strength ofepistasis was generally weak (most interactions resemble a parallelogram, Fig 3.3 and Fig 3.4). Significantsign epistasis was only observed in a single diploid case (erg3 erg7).Epistatic interactions in YPD were qualitatively different from those observed in nystatin2 and oftendiffered between haploids and diploids (Table 3.4 and Table 3.5). In contrast to the prevalence of negativeepistasis in nystatin2, significant positive epistasis was observed in some cases (the double mutant is morefit than expected under the additive model). The poor growth in YPD of most double mutant strains suggeststhat the negative relationships observed in nystatin2 may, in part, be due to intrinsic growth defects, perhapsdue to the instability of the cell membrane without proper ergosterol synthesis.3.3.4 Tolerance across a range of nystatinTo see whether the genetic interactions depended on the concentration of drug, growth was measured asoptical density (OD) after 24 hours over a range of nystatin concentrations (0, 1, 2, 4, 8, 16, 32, 64, 128, 256µM).We focused here on OD to assess the range of environments in which the yeast strain could grow, even ifslowly, and because of the massive replication required. While OD is thought to reflect the efficiency of cells’603.3. ResultsTable 3.4: Results from a mixed-effects model run on all genes using the haploid maximum growth rate datain YPD. For statistical and column details, see Table 3.2. There were no cases of sign epistasis.Term Coefficient SE P Epistasismating type 0.000042 0.0027 0.98erg3 -0.029 0.0057erg5 -0.0051 0.0049erg6 -0.065 0.0049erg7 -0.12 0.0050erg3*erg5 -0.026 0.0091 0.0034 negativeerg3*erg6 0.0030 0.0090 0.74erg3*erg7 0.018 0.0091 0.041a positiveerg5*erg6 0.0018 0.0077 0.81erg5*erg7 -0.0065b 0.0079 0.41erg6*erg7 0.040 0.0084 1.77 ⇥ 106 positiveaNot significant when outliers are included.bPositive when outliers are included.Table 3.5: Results from a mixed-effects model run on all genes using the homozygous diploid maximumgrowth rate data in YPD. For statistical and column details, see Table 3.2. There were no cases of reciprocalsign epistasis.Term Coefficient SE P Epistasis Signerg3 -0.057 0.011erg5 -0.076 0.010erg6 -0.14 0.010erg7 -0.19 0.010erg3*erg5 0.020 0.022 0.36erg3*erg6 0.032 0.022 0.14erg3*erg7 0.15 0.022 8.5 ⇥ 1011 positiveerg5*erg6 0.079 0.020 5.6 ⇥ 105 positiveerg5*erg7 0.056 0.018 0.0021 positiveerg6*erg7 0.016 0.025 0.53613.3. Resultsability to turn nutrients into cellular material rather than the rate of growth, OD and maximum growth ratewere correlated for the single mutants analysed here (GERSTEIN et al. 2012), and the interactions observedwere qualitatively similar for the concentrations of nystatin used in both the maximum growth rate and ODassays (0 mM and 2 mM).As before, mating type was not found to have a significant effect on OD in the haploid data (linear modelthat included mating type, concentration of nystatin and strain identity as fixed effects; mating type: F =0.23, df = 1, P = 0.63; concentration of nystatin: F = 600.12, df = 1, P < 1015; strain: F = 31.95, df = 10,P < 1015), and data were pooled across mating types.Figure 3.5: Optical density after 24 hours of growth for haploid strains in a range of concentrations ofnystatin. Colours go from red to purple, through blues, from lowest to highest concentrations of nystatin.Lines connect different mutants in the same concentration of nystatin. Lines are solid when the difference inOD is significant in a Welch’s t-test and dotted when non-significant (not adjusted for multiple comparisons).Arrows on the y-axes indicate the OD of the ancestral strain. Error bars denote the standard error acrossreplicates. All underlying raw data and analyses can be found in ONO et al. (2016).We found that the form of gene interactions changed when measured over a range of concentrationsof nystatin (haploid results: Fig 3.5). As observed previously, the double mutant generally had equivalentor lower growth than the two parent mutants at low concentrations of nystatin (0-4 µM), but at high con-centrations (32-64 mM), the double mutant strains became the only strains able to grow well. That is, apreponderance of negative epistasis shifted towards a preponderance of positive epistasis as nystatin con-centrations rose. This dependence of the sign of epistasis on the concentration of the drug (not only on thepresence or absence of the drug) indicates that the outcome of mutation or hybridization will depend heavily623.3. Resultson the specifics of the environment in which the yeast is found.Homozygous diploid strains showed qualitatively similar patterns of growth to the haploid strains, withthe exception of the erg6/erg6 erg7/erg7 double mutant (Fig B.3). When we compared all diploid strains (in-cluding heterozygous strains), interesting patterns emerge (Fig B.5). In many cases, the double heterozygousstrain exhibited more growth than either single heterozygous strain (as observed by a ‘bump’ in the middleof the figure), particularly at higher concentrations of nystatin. This may indicate a net beneficial effect ofcarrying two heterozygous mutations or may reflect an increased potential for loss of heterozygosity (LOH).LOH, in which a locus that is initially heterozygous for a mutant allele becomes homozygous, would bebeneficial in our fitness assay because being homozygous for either mutant allele increases growth in nys-tatin (compare middle point in Fig B.5 to those second from either end). This may have occurred duringthe course of the fitness assay, affecting our final measures of fitness. LOH was previously observed forthe single heterozygous mutants over a 72-hour time scale (GERSTEIN et al. 2014), and being heterozygousfor two mutations may increase the chance of LOH for at least one of the two. The unexpected increase infitness in the double heterozygotes may also be indicative of an epistatic interaction providing some benefitto having two heterozygous mutations within the ergosterol pathway compared to full recessivity (i.e., nobenefit) with only a single heterozygous mutation (GERSTEIN et al. 2014).3.3.5 Ergosterol phenotypes and map to fitnessTo determine whether epistasis for fitness was consistent with the sterol phenotypes exhibited by the strains,we extracted and measured the sterol profile of allMATa strains. In ancestral samples, we see the characteris-tic four-peaked curve between 240 and 300 nm that is produced by ergosterol and the late sterol intermediate24(28)dehydroergosterol (ARTHINGTON-SKAGGS et al. 1999). Only the latter sterol shows an absorptionband at 230 nm, allowing quantification of ergosterol, but we found the peak between 200 and 230 nm to bevery sensitive to the standard used (e.g., newly mixed heptane and ethanol vs. heptane layer from extractionperformed with no yeast cells and ethanol) and thus limit ourselves to a qualitative description of the results.All of our single mutants show similar results to those presented by GERSTEIN et al. (2012) for thesesame mutants (Fig 3.6). The two potential loss-of-function mutants (erg3 and erg6) also have similar sterolprofiles to knockout mutants of these genes (JENSEN-PERGAKES et al. 1998; MUKHOPADHYAY et al. 2002).Double mutants show a variety of profiles, as can be seen in Fig 3.6. Notably, most double mutants resembleone of the two parent single mutants, with the exception of the erg6 erg7 double mutant, which is interme-diate between the two single mutants in absorbance over much of the measured range (suggesting a mixtureof sterols present). All double mutants that include the mutation in ERG3 tend to show similar profilesto the erg3 single mutant. Thus, the sterol profiles were not predicted by gene position in the ergosterolbiosynthesis pathway (as ERG6 is upstream of ERG3). Furthermore, the similarity in sterol profiles betweendouble and single mutants did not generally predict the patterns observed for maximum growth rate (with theexception of the erg5 erg7 haploid and diploid, which behaved like erg7, and the erg3 erg7 diploids, whichbehaved like erg3), indicative of a disconnect between sterol profile and fitness.633.4. DiscussionFigure 3.6: Sterol profiles of all MATa haploid strains as measured using a spectrophotometry-based assay.The colour scheme is the same as in Fig 3.3, with the double mutant in black and the ancestral strain in grey.Error bars depict the standard error of three replicates with the exception of erg6 erg7 (2 replicates). Thesame ancestral and single mutant assays are represented in multiple panels. All underlying raw data andanalyses can be found in ONO et al. (2016).3.4 DiscussionWe investigated the types of genetic interactions present between pairs of first-step beneficial mutations thatarose independently in the presence of the fungicide nystatin. We focused on four mutations, representingeach gene found to carry a beneficial mutation among 35 strains evolved in 4 mM nystatin (GERSTEIN et al.2012). All of these genes are in the biosynthesis pathway leading to the production of ergosterol (the pri-mary sterol in the yeast cell membrane, Fig 3.2). When ergosterol is bound by nystatin, the cell membranebecomes permeable to ions, sugars, and metabolites (CARRILLO-MUÑOZ et al. 2006), and cell death results.When assayed at 2 mM nystatin, the interactions found among these beneficial mutations were predominantlynegative, with double mutants exhibiting a lower growth rate in nystatin than expected based on the com-bined benefits of the single mutations. This negative epistasis was observed in both haploids (Fig 3.3) andhomozygous diploids (Fig 3.4), supporting previous findings that interactions between mutations in func-tionally related genes are often negative (COSTANZO et al. 2016).3.4.1 Prevalence of sign epistasisWe find that the interactions were so negative that the double mutant grew less well than at least one ofthe parent single mutants (sign epistasis) in four of the six gene combinations assayed in haploids. In halfof these cases, the double mutant grew significantly less well than both single mutants (reciprocal signepistasis). Similar interactions were observed in diploids (three cases of sign epistasis, two of which werereciprocal). The observation of reciprocal sign epistasis is of particular interest, as this type of BDM in-compatibility underlies postzygotic reproductive isolation among speciating lineages. The high frequency ofreciprocal sign epistasis observed, even among first-step beneficial mutations acquired in the same environ-ment, confirms the possibility that isolated populations experiencing similar selective pressures can diverge643.4. Discussionand eventually speciate simply through the order of mutations that happen to arise and fix (mutation-orderspeciation).3.4.2 Maximum growth rate in one environment does not predict sterol phenotype orgrowth in other environmentsThe prevalence of sign epistasis among our specific set of beneficial mutations is somewhat surprising giventhe linearity of the biosynthetic pathway in which all of the affected ergosterol genes act (Fig 3.2). Ourresults were not consistent with the expectation that the phenotype and fitness of double mutants would bedetermined by the upstream mutation. In terms of phenotype, the sterol profile of the double mutant wassimilar to that of the most upstream mutant in only two cases (the erg5 erg7 and erg3 erg5 double mutants,Fig 3.6). In terms of fitness, the growth rate of the double mutant differed significantly from that of the mostupstream single mutant in three (haploids) and four (diploids) out of six pairwise comparisons (Figs 3.3 and3.4).In the combination of two loss-of-function type mutations (erg3 erg6), neither sterol phenotype nor fit-ness matches that of the upstream mutation. These results indicate that there remain substantial interactionsbetween the mutations in the ergosterol pathway, potentially due to partial activity of the upstream genes cre-ating low levels of substrate for the remainder of the pathway, due to downstream genes acting on alternativesterol substrates, or due to interactions among the intermediate sterols themselves. From previous work inthe yeasts S. cerevisiae and Candida albicans, it has been shown that ERG6 plays a role in offshoot sterolsynthesis in mutants of ERG3 (SANGLARD et al. 2003), and it is known that intermediate sterols are found atdifferent levels in different compartments of the cell (ZINSER et al. 1993) and may impact fitness in a varietyof ways (e.g., altering temperature tolerance CASPETA et al. 2014 and virulence MCCOURT et al. 2016).There was also no clear relationship between sterol phenotype and fitness in these strains. Sterol pheno-type for most double mutants resembles one of the two single mutants (Fig 3.6), but this similarity in sterolphenotype did not generally predict maximum growth rate in nystatin2 (with the possible exception of erg5erg7 in haploids and diploids and erg3 erg7 in diploids, Figs 3.3 and 3.4). Future analyses that determinethe processing of sterols in the single and double mutants, as well as their pleiotropic effects, would furtherelucidate these genetic interactions.Interestingly, the type of epistasis depended strongly on the concentration of nystatin. At lower concen-trations of nystatin, similar to those used to acquire the mutations (4 mM nystatin), epistatic interactionswere typically negative (Fig 3.5), with the double mutant showing similar or lower densities after 24 hoursof growth than the single mutants. By contrast, at higher concentrations of nystatin, the interactions wereoften positive, with double mutants typically able to outgrow both single mutants. Emblematic of this phe-nomenon, the best growing haploid double mutant strains at 32 mM nystatin (erg3 erg6, erg3 erg7, erg6 erg7)were also those that exhibited the most negative epistasis at lower concentrations. This implies a tradeoffbetween growth in a lower concentration of the fungicide and tolerance to high concentrations of the drug.Conceptually, this tradeoff suggests that the double mutant initially overshoots the optimum when nystatinconcentrations are low, because the costs associated with each ergosterol mutation are combined (perhapsdestabilizing the plasma membrane); by contrast, when nystatin concentrations are high, the optimum is653.4. Discussionshifted even further away, and extreme reductions in ergosterol and potentially other sterols are needed forthe yeast to survive, at which point the double mutant is most fit (see, e.g., BLANQUART et al. 2014 for atheoretical exploration of this phenomenon). Because membrane damage can trigger cell cycle arrest in yeast(KONO et al. 2016), another possible explanation for the results observed at high concentrations of nystatinis that single mutants experience cell cycle arrest, reducing growth rate, whereas the additional stress causedby the combination of two mutations and high concentrations of nystatin may cause a checkpoint failure indouble mutants, allowing the cells to bypass arrest and continue dividing (C. Nislow, pers. comm.).The shifting nature of epistasis as a function of the severity of the environment also has implications forspeciation and has not been widely discussed (but see KISHONY and LEIBLER 2003 for discussion aboutenvironment-dependent epistasis and ARNEGARD et al. 2014 for an example of an environment-dependentnegative epistatic interaction on feeding and growth performance in F2 hybrid stickleback). Our results showthat BDMs can be environment-specific, and thus gene flow between species might vary according to theenvironment in which secondary contact occurs (BORDENSTEIN and DRAPEAU 2001). Counterintuitively,our results further suggest that harsher environments may be more conducive to gene flow because of thepossible benefit of combining adaptive mutations from different populations. Indeed, environments thatare so harsh that only strains combining mutations can survive (as we observed at high concentrations ofnystatin) might promote hybridization and potentially lead to hybrid speciation (reviewed in MALLET 2007).For example, extreme desert environments have selected for combinations of traits that improve droughttolerance, allowing hybridHelianthus sunflowers to colonize and proliferate (GROSS and RIESEBERG 2005).3.4.3 Fitness landscapes in haploids and diploidsBecause cell volume to surface area ratios are different for haploids and diploids (MABLE 2001), we mightexpect differences in growth and epistasis between haploids and diploids, particularly in the face of a selec-tive pressure like nystatin that impacts the cell membrane. By and large, however, our results were consistentregardless of ploidy, with diploid homozygous mutants and haploid mutants showing similar patterns of epis-tasis. One exception was the erg6/erg6 erg7/erg7 double mutant, which was so unfit in diploids that yieldswere often too low to obtain initial cell densities similar to other strains in our growth rate assays, evenafter extended incubation. The haploid version of this same double mutant, however, also showed very lowfitness. During the initial isolation of the haploids from spores, the double mutant colonies were identifiableby their noticeably smaller size compared to those produced by single mutant and ancestral genotypes. Ahaploid double mutant strain also exhibited reversion in one instance during growth in 10 mL YPD (Sangersequencing revealed a secondary mutation in the same codon as the original mutation, reverting the aminoacid).Considering the various diploid heterozygotes, we confirmed that the ergosterol mutations were largelyrecessive, as found previously for the single heterozygous mutant strains (GERSTEIN et al. 2014). Therewere more signs of nystatin resistance in the double mutant strains than in the single mutants, however. Oneindication of this was the double heterozygous strains showing a slight increase in biomass produced (asmeasured by optical density) compared to the single heterozygous strains across a range of concentrationsof nystatin (Fig B.5). Despite this, the double heterozygous strains were uniformly of low growth rate in663.4. Discussionnystatin2 (open diamonds in Fig 3.4), with similar sensitivity as found in the ancestor. The generally poorperformance of the double heterozygous diploid is of particular interest because this genotype would be thefirst hybrid product of crosses between strains fixed for different beneficial mutations (see also Fig B.4).Thus, F1 hybrid inviability in the double heterozygotes, as well as reciprocal sign epistasis, contributes toreproductive isolation between these strains.3.4.4 Implications for speciationOverall, we find that the very earliest stages of divergence within a common selective environment can gen-erate postzygotic reproductive isolation, observing sign epistasis, reciprocal sign epistasis, and F1 hybridinviability in double heterozygotes among the first-step adaptive mutations isolated in the presence of nys-tatin. Although we did not assay incompatibilities at other stages (e.g., meiotic incompatibilities), we expectthat further BDMs might be revealed by analyzing other stages in the life cycle (indeed, it was very difficultto sporulate some double mutant strains, particularly erg5/ERG5 erg7/ERG7). We speculate that geneticincompatibilities may be especially likely in scenarios such as the one investigated here, where selectionfavours large effect mutations. In the initial experiment in which mutations were acquired, the concentrationof nystatin was chosen to inhibit growth, so that only mutations capable of rescuing fitness were isolated(GERSTEIN et al. 2012). Such large effect mutations might have more costly pleiotropic effects and/or bemore likely to overshoot the fitness optimum when combined, showing negative epistasis for fitness even iftheir effects are multiplicative or additive on the underlying trait. If large effect mutations are more likelyto interact negatively, which is consistent with our results and others (CHOU et al. 2011; KHAN et al. 2011;KRYAZHIMSKIY et al. 2014; SCHENK et al. 2013), short periods of severe selection might be more likelyto lead to speciation than longer periods of mild selection. Future experiments comparing genetic incom-patibilities among strains with similar levels of divergence but consisting of a few large effect or severalsmall effect genetic differences would be extremely valuable. We also speculate that independent popula-tions experiencing directional selection to the same environmental change might be more likely to speciatethan those experiencing directional selection to different environments because the beneficial mutations thataccumulate in the former case may be more likely to involve similar pathways and thus more likely to in-teract negatively (as has been shown in interaction studies, see COSTANZO et al. 2016; HARTMAN et al.2001). Indeed, even though the beneficial mutations that we assayed were all in the same ‘linear’ pathwayand acquired in the same selective environment, we found that the type of epistasis that underlies speciationwas common, providing experimental support for the mutation-order speciation hypothesis.67Chapter 4The limit to evolutionary rescue depends onploidy in yeast exposed to nystatin4.1 IntroductionAs their environment changes, species must adapt to persist. By knowing the factors that affect populationsize and adaptation, we can begin to predict or influence a population’s likelihood of survival. Evolutionaryrescue occurs when a population is saved from eventual extinction in response to an environmental changeby genetic adaptation to this new environment (CARLSON et al. 2014). Broadly, evolutionary rescue is im-portant to understand from two main standpoints. The first is conservation - where the question is whethera species will be able to survive current and future environmental changes. The second is related to humanhealth - where the question is whether pests and pathogens are able to evolve to overcome human interven-tion. In evolutionary rescue, populations increase their growth rate either using existing standing geneticvariation, new alleles introduced by gene flow, or new mutation. Populations reach their adaptive limit whenthey are not able to acquire the genetic changes necessary for rescue before going extinct. In this paper,we focus on the case of evolutionary rescue from new mutation, asking: Does ploidy influence the limits toadaptation?The genomic characteristics of the organism may influence the potential for evolutionary rescue in a pop-ulation. For example, epistasis between potential adaptive mutations and the genetic background can affectwhich mutations are actually beneficial. Here, we focus on ploidy as a major genomic factor that can impactthe outcome of evolution. For a variety of reasons, diploid and higher ploidy populations are expected to havedifferent rates of evolutionary rescue from haploid populations (OTTO and WHITTON 2000). With diploidy,the dominance of adaptive mutations can strongly influence the likelihood of rescue because recessive mu-tations will not be ‘seen’ by selection when they initially arise in heterozygous form. Recessive mutationswill then have a smaller chance of rising to high frequency in the population and will take longer to do socompared to partially dominant or fully dominant mutations (ORR and OTTO 1994). However, many pestsand pathogens are either haploid (for example, infectious bacteria), alternate between haploid and diploidphases or have haploid and diploid individuals (for example, the spider mite pest Tetranychus urticae hashaploid males and diploid females). For these organisms, recessive mutations can be immediately beneficialif they have a positive selection coefficient in the haploid individuals/phase and can aid in the evolution ofantibiotic or pesticide resistance. On the other hand, diploids should acquire twice the number of mutations,if haploids and diploids have the same per basepair mutation rate, which may allow diploids to be rescuedwhen mutations are sufficiently dominant (e.g., OTTO and WHITTON 2000), especially when adaptation is684.1. Introductionlimited by mutational availability (ORR and OTTO 1994). Experiments performed using yeast support thisprediction. In large populations, adaptation is faster in haploids because mutations are not limiting, but whenpopulation sizes are decreased, the haploid advantage is lost (ZEYL et al. 2003).The type of environmental change also determines the ease of evolutionary rescue from new mutation. Ifa higher fraction of possible mutations are adaptive in the new environment, evolutionary rescue should occurmore frequently because any individual mutation will have a higher chance of being beneficial. A relatedfactor may be the rate of environmental change. In experimental populations of Escherichia coli adaptingto the antibiotic rifampicin, LINDSEY et al. (2013) determined that rapidly changing environments not onlylimit the number of available mutations by lowering population size but also make certain evolutionarytrajectories inaccessible because multiple mutations are required, and these mutations are not all individuallybeneficial in high concentrations of the drug. As a result, there is a much smaller fraction of survivingpopulations when there are faster rates of environmental change.Little is known about the effect of ploidy on evolutionary rescue from new mutations and the topic hasnot been well-discussed in the theoretical literature (reviewed in ALEXANDER et al. 2014), thus makingit an interesting avenue of exploration. Most investigations have focussed instead on factors such as therate of population size decline with environmental stress and the relative contribution of new mutations vs.standing genetic variation. Some models of evolutionary rescue are designed with or can be extended todiploid populations but they tend to assume that the dominance of rescuing mutations is relatively high,giving the heterozygote a fitness advantage over the ancestral type (e.g., GOMULKIEWICZ and HOLT 1995,ORR and UNCKLESS 2008). When evolutionary rescue is allowed to occur from standing genetic variation,as opposed to new mutations, dominance is not as important because the necessary mutation may alreadybe present at a high enough frequency to make homozygotes common (ORR and UNCKLESS 2008). Infact, if the rescue mutation was deleterious prior to the change in environment and populations are startedfrom mutation-selection balance, a recessive allele is as likely to fix as a dominant allele because morecopies of the recessive allele will already be present in the population (ORR and BETANCOURT 2001).If standing genetic variation is produced through hybridization between the focal population and either adivergent population or a closely related species, diploidy may even be favoured if beneficial mutationsare often dominant or overdominant (as observed between species of yeast, BERNARDES et al. 2017) anddeleterious mutations or incompatible alleles between groups are recessive (predicted to be the case betweenyeast species based on STELKENS et al. 2014). Less is known about the dominance of beneficial mutationsin general, however, which is important for populations of a diploid organism relying on new mutations forevolutionary rescue. Much of the relevant experimental work has been done with bacteria (e.g., studies ofantibiotic resistance evolution reviewed in MACLEAN et al. 2010) or haploid eukaryotes (e.g., haploid yeastin BELL and GONZALEZ 2009 or primarily haploid Chlamydomonas reinhardtii in LACHAPELLE and BELL2012), thus a study of the effects of ploidy on evolutionary rescue will bring novel insights.We investigate the ability of yeast, Saccharomyces cerevisiae, to undergo evolutionary rescue in responseto a high concentration of the fungicide nystatin and how the ploidy of the yeast affects this ability. We modelour design after a previous experiment in this system performed exclusively with haploids (GERSTEIN et al.2012), which found that the adaptive alleles acquired by haploid yeast were all recessive in the diploid state694.2. Methods(GERSTEIN et al. 2014). These recessive alleles will not be sufficient to rescue diploid populations of yeast,as they will appear in heterozygotes, but the mutations found by GERSTEIN et al. (2012) are unlikely to bethe only possible beneficial mutations in this environment, potentially missing dominant beneficial alleles.Diploids, aided by their larger genome size and therefore higher number of mutations per cell, may explorea wider range of potentially adaptive mutations, finding some that are at least partially dominant. For thesereasons, we expected diploid yeast to access alternative evolutionary paths to the haploids, albeit diploidswould likely undergo evolutionary rescue less frequently due to the apparently low availability of dominantbeneficial mutations in this environment. Supporting this idea is a similar study performed by ANDERSONet al. (2004) that evolved both haploid and diploid yeast to the drug fluconazole, for which candidate path-ways to drug resistance are known. At low concentrations of the drug, mutations in PDR1 and PDR3 werefavoured, and diploids were able to adapt faster than haploids due to increased mutation availability andtherefore decreased waiting time. All mutations found in diploids were dominant (29/29) while only abouthalf of those found in haploids were dominant when tested in a heterozygous diploid background (12/29) de-spite almost all being found in one of those two genes. At high concentrations of the drug, however, recessivemutations in ERG3 (one of the genes also implicated in nystatin resistance) are favoured, and diploids werefound to evolve slower than haploids (ANDERSON et al. 2004). The authors concluded that diploids likelyrequired two mutational events to occur (a mutation in ERG3 followed by a second mutation that renderedthe first mutation in ERG3 homozygous) in order to acquire resistance, slowing their adaptation. In the cur-rent study, we set out to determine the mutations involved in the evolutionary rescue of diploid populationsto a high concentration of nystatin and found that, by and large, diploid populations did not genetically adaptwithin our short time course evolution experiments, even though we observed hundreds of cases of rescue inhaploid populations over this same time period. These results have implications for the efficacy of nystatinwhen applied to fungal pathogens, which include Candida albicans, a common human fungal pathogen thatis predominantly diploid (HICKMAN et al. 2013), and S. cerevisiae, for which clinical strains are mostlyeither diploid or of higher ploidy (ZHU et al. 2016) .4.2 Methods4.2.1 StrainsIn total, three mutant acquisition experiments, similar to GERSTEIN et al. (2012), were performed, plusone flask experiment conducted at large population size. Except where noted, we used the S288C back-ground, using the strains BY4741 (MATa his31 leu20 met150 ura30), BY4739 (MAT↵ leu20lys20 ura30) (Open Biosystems) and a diploid produced by mating the two (BY4741xBY4739, mat-ing described in ONO et al. 2017). To assess the sensitivity of the results to strain background, Acquisi-tion Experiment 3 was performed using a different background (W303), using the haploid strains MJM64(MATa-YCR043C::KANMX STE5pr-URA3 ade2-1 his3::3xHA leu2::3xHA trp1-1 can1::STE2pr-HIS3STE3pr-LEU2) and MJM36 (MAT↵-YCR043C::HPHB STE5pr-URA3 ade2-1 his3::3xHA leu2::3xHAtrp1-1 can1::STE2pr-HIS3 STE3pr-LEU2) constructed by MCDONALD et al. (2016), and a diploid producedby mating (OLY075). The diploid was generated following the procedures of MCDONALD et al. (2016) with704.2. MethodsFigure 4.1: Visual representation of all mutant acquisition experiments. The first two experiments wereperformed in 96-well deep well boxes using the BY strains (BY4741, BY4739 and BY4741xBY4739) andYPDnystatin4 as the medium. For two wells from Acquisition Experiment 1 and three wells from Acqui-sition Experiment 2, the protocol was not followed correctly, and these wells are excluded from the paperentirely. The third acquisition experiment was performed similarly except that it usedW303 strains (MJM64,MJM36 and OLY075) and SCnystatin4. In this experiment, due to space constraints, 80 wells were inocu-lated from the same pre-growth culture as another well in the experiment (all MAT↵). In the analysis, werestrict ourselves to considering only one of these two replicates because they are not independent. If neitherreplicate grew, we counted that as no growth (63 cases). If one of the two replicates grew, we counted thatas growth and performed further assays on the population that grew (13 cases). If both replicates grew, wechose one at random to analyze and discarded the other (4 cases). The final acquisition experiment inves-tigated a much larger population size (roughly 100-fold greater), being performed in flasks instead of deepwell boxes and again using the BY strains and YPDnystatin4.slight modifications; using YPAD in place of YPD (YPD + 40 mg/L adenine sulfate), incubating the matingsovernight, excluding the PBS buffer step, and performing the selection step twice. The growth of all strainswas inhibited under the treatment conditions, with no observable positive growth in the absence of a resistantmutant within 48 hours (Fig. C.2).4.2.2 Mutant acquisition in deep well boxesFor an overview of all acquisition experiments performed, see Fig. 4.1. For Acquistion Experiments 1 and2, the strains were first struck from frozen on YPD plates and grown at 30°C. After three days of growth,100-well honeycomb plates used with the Bioscreen C Microbiological Workstation (Thermo Labsystems)were filled with 150 mL of YPD per well, and each well was inoculated with a separate colony to producestationary phase cultures for use in the mutation acquisition phase. The plates were grown for 24 hours in theBioscreen machine at 30°C with maximum, continuous shaking. Note that all liquid medium throughout thispaper was supplemented with ampicillin at a final concentration of 0.04 mg/mL to prevent bacterial growth.In Acquisition Experiment 1, we used 191 colonies of the diploid strain and 191 colonies of the haploidstrains, split between mating types (MATa: 96, MAT↵: 95), alternating between diploid and haploid strainsthroughout the plates. In Acquisition Experiment 2, we used 286 colonies of the diploid strain, 47 coloniesof MATa, and 48 colonies of MAT↵, with any given plate containing either all diploids (three plates) or halfMATa and half MAT↵ (one plate).714.2. MethodsThe following day, the honeycomb plates were visually assessed to confirm full growth in each wellwithin YPD. 10 mL from each well was used to inoculate 990 mL of YPD + 4 mM nystatin (‘YPDnystatin4’)in deep well boxes, after mixing the culture by pipetting up and down. We estimate this inoculum to contain~7.0⇥ 105 cells per well (based on hemacytometer counts). The same general map was used for the boxesas for the honeycomb plates. The boxes were covered with aluminum lids to prevent cross-contaminationwhile sampling and plastic lids were added on top to protect the aluminum lids. They were incubated at30°C, shaking at 200 rpm.The wells were checked every 24 hours for growth by visual examination. Clumps were typically ob-served at the bottom of the well prior to the yeast covering a larger, circular area, so a well was consideredto have growth if the bottom was covered with enough yeast cells to form two small clumps, about 2 mmlong and 1 mm wide. On the first day that growth was observed in a well, the growth was recorded, thealuminum lid was sterilized with 70% ethanol and punctured with a pipet tip, and the culture was frozenin 15% glycerol at -80°C. In Acquisition Experiment 1, 42 MATa strains, 86 MAT↵ strains and 90 diploidstrains were collected over the course of 12 days. In Acquisition Experiment 2, 16 MATa strains, 48 MAT↵strains and 100 diploid strains were collected over the course of 10 days.Acquisition Experiment 3, performed in the W303 background, used similar methods to AcquisitionExperiment 1, with exceptions described here. The strains were originally struck from frozen on YPADplates and grown for only two days. SC (supplemented with adenine) was used instead of YPD for growth inthe honeycomb plates. SC was formulated using 20 g/L of dextrose, drop-out mix complete (US Biological,D9515), and yeast nitrogen base including ammonium sulfate, according to the manufacturer’s instructions.This medium was supplemented with an additional 57 mg/L of adenine sulfate. There were 155 colonies ofthe diploid strain, 80 colonies of the MATa strain and 165 colonies of the MAT↵ strain used for the growthin the honeycomb plates, alternating between strains throughout. SC + 4 mM nystatin (‘SCnystatin4’) wasused in place of YPDnystatin4 in the deep well boxes. Initial inoculum was estimated as 7.4⇥ 105 cells perwell. Pilot experiments indicated that mutants would be difficult to isolate in the MAT↵ background (theinitial stock was later found to be respiratory-deficient), so 80 additional wells of this strain were added tothe deep well boxes, using the same pre-growth culture as for one other well. A total of 245 wells of theMAT↵ strain were included in the experiment, so the map of the deep well boxes was slightly modified fromthat of the honeycomb plates, although still alternating between strains. Over the course of seven days, 77MATa strains, 34 MAT↵ strains and 121 diploid strains were collected.4.2.3 Confirming nystatin resistanceAll populations frozen from the acquisition experiments were tested for resistance to nystatin. Populationswere pre-grown from frozen by inoculating 975 mL of 0.5 mM nystatin (in YPD for Acquisition Experiments1 and 2, SC for Acquisition Experiment 3) with 25 mL of frozen culture in deep well boxes according to arandomized map. Aluminum lids were added to the boxes, and they were incubated at 30°C, shaking at 200rpm for 72 hours, after which almost all wells had full growth as judged by visual inspection. From thesepre-growth cultures, 200 mL was transferred to a 1.5 mL tube and stored at 4°C.We measured growth in YPDnystatin4 (for Acquisition Experiments 1 and 2) or SCnystatin4 (for Acqui-724.2. Methodssition Experiment 3) of all populations on three different days using the Bioscreen C, which automaticallymeasures optical density (‘OD’) in 100-well honeycomb plates. For each growth assay, honeycomb platesfilled with 148.6 mL of nystatin medium were inoculated with the cultures kept at 4°C according to a newrandom map for each day. The culture tubes were vortexed until fully resuspended, and 1.5 mL was trans-ferred to the appropriate well. All tubes were returned to 4°C when inoculation was completed. OD wasmeasured automatically using the wideband filter at 30-min intervals for 72 hours from cultures growing at30°C with medium continuous shaking.Lines were considered potentially mutant when the well had an OD after 72 hours of growth (‘OD72’)that was greater than the halfway point between the well with the lowest OD and that with the highest OD,including the ancestral controls, in the majority of assays (at least two out of three). We consider these to be‘potentially’ mutant because growth of the yeast can depend on the exact conditions of the assay (includingplate type and volume of medium), thus they are not necessarily fully resistant to the original evolutionaryconditions (deep well box, 1 mL of medium). Based on initial testing, we knew that our collection of haploidpopulations included some nystatin-resistant mutants, so the highest OD reflected a truly resistant strain. Asingle cutoff was used for both Acquisition Experiment 1 and 2 for each assay day since they were tested inthe same medium and always assayed together. Because Acquisition Experiment 3 used a different mediumand the assays were conducted on different days, the cutoffs were recalculated.4.2.4 Mutant acquisition with larger population sizesBecause we had such difficulty finding diploid mutant strains using relatively small population sizes (roughly7⇥ 105 cells inoculated) in our original acquisition experiments, we conducted a mutant acquisition experi-ment with a higher population size (Fig. 4.1). To do so, we first struck BY4741, BY4739 and BY4741xBY4739from frozen onto YPD plates and allowed them to grow for three days at 30°C. We then inoculated 10 mLof YPD in separate test tubes for colonies from 10 diploid colonies and one of each haploid mating type.These tubes were allowed to grow for 24 hours in a rotor at 30°C. The next day, we inoculated 250 mL flasksfilled with 99 mL of YPDnystatin4 with 1 mL of the overnight culture (12 flasks in total). We estimate thisinoculum to contain ~7.0⇥ 107 cells (based on counts in a hemacytometer).The flasks were covered with aluminum foil and incubated at 30°C, shaking at 200 rpm. The flaskswere checked every 24 hours by visual examination for growth. When growth was observed (as a noticeablelightening of the culture colour and loss of clarity when compared to a flask containing no yeast), 500mL of culture was sampled and frozen at -80°C in 15% glycerol. All flasks showed growth by Day 10.Resistance to nystatin was then assayed as described for the first two acquisition experiments with cutoffsbeing recalculated because these assays were performed on different days.4.2.5 Further testing of potential diploid mutantsAll 13 diploid populations that consistently showed growth in nystatin and underwent further testing hadthe BY genetic background since all were from Acquisition Experiments 1 or 2. Because initial resultsindicated that some ‘diploid’ mutants were actually haploid contaminants, all potential diploid mutants wereverified by replica plating. Haploids of this background carry auxotrophies and can thus be detected both734.2. Methodsby their ability to mate with haploids of the opposite mating type (tested by mating with haploid strainscarrying different auxotrophic mutations and testing for complementation) and by their inability to grow oneither SC lacking histidine (MATa) or SC lacking lysine (MATa). In this way, we determined that nine ofthe populations were primarily composed of haploid contaminants, and these were excluded from furtheranalyses. Four diploid potential mutant populations remained after these tests.These four diploids, along with ancestor and haploid mutant controls, were further tested for growth inYPDnystatin4. Yeast were struck from frozen on YPD plates and incubated at 30°C for three days. Betweensix and eight colonies per population were chosen haphazardly, picked into 1 mL of YPD in deep well boxes,and incubated at 30°C, shaking at 200 rpm overnight. From this culture, 10 mL was used to inoculate 990mL of YPD and another 10 mL was used to inoculate 990 mL of YPDnystatin4, both in deep well boxes.Aluminum lids were added to these boxes to minimize evaporation. The boxes were incubated at 30°C,shaking at 200 rpm for 72 hours and visually inspected for growth every 24 hours. Growth was scored ona scale from zero (no growth) to three, where one is barely recognizable growth (a light dusting of cellsor a circle of roughly 1 mm diameter), two is recognizable but not full growth (a patchy covering of cellsor a circle up to roughly 3 mm diameter) and three is full growth (a circle of greater than roughly 3 mmdiameter up to a full covering of the bottom of the well). Wells were chosen for freezing based on theirgrowth after the 72 hour incubation. For the populations that showed full growth in YPDnystatin4, one wellwas chosen randomly for freezing (coming from a single colony). If a population showed variable growthacross colonies, the colony showing the best growth was chosen for freezing, or one of the colonies showingthe best growth was chosen randomly. Colonies were frozen from the first YPD box in 15% glycerol at-80°C.A growth assay was performed using the Bioscreen C machine on the colonies chosen for freezing, alongwith randomly chosen colonies from the strains that did not show growth in YPDnystatin4, using culture fromthe second YPD box (inoculated on the same day as the YPDnystatin4 box). 1.5 mL of culture was used toinoculate 148.6 mL of both YPD and YPDnystatin4 in four wells each, using freshly diluted nystatin stocks.OD was measured automatically using the wideband filter at 30-min intervals for 72 hours from culturesgrowing at 30°C with medium continuous shaking. No diploid colony showed growth in this growth assay,and it was later determined that these four diploid populations also harboured haploid contaminants at verylow frequencies. This was determined by testing the culture from the end of the original growth assays asdescribed at the beginning of this section.4.2.6 Nystatin efficacy over timeWhile many diploid cultures appeared to undergo evolutionary rescue in nystatin, growth was typicallyonly observed late in the acquisition experiment, and most of these populations did not exhibit resistancewhen regrown in nystatin. We thus tested the ability of nystatin to inhibit yeast growth over a time course toinvestigate whether the effective concentration of the drug was decreasing over the duration of the acquisitionexperiments. We also tested whether the presence of dead yeast cells altered the efficacy of nystatin. Deadcells were obtained by growing cultures of BY4741xBY4739 in 1 mL of YPD in 1.5 mL tubes overnight at30°C, shaking at 200 rpm. The next day, these tubes were placed in a heat block at 95°C for 10 minutes to744.3. Resultskill the cells. Preliminary tests indicated that this amount of time was sufficient to kill all of the cells. On thefirst day of the experiment (Day 0), 990 mL of YPD, YPDnystatin4, SC and SCnystatin4 were placed in deepwell boxes, with half of all nystatin-containing wells also receiving 10 mL of heat treated cells, giving a totalof 16 wells per type (base medium, base medium with nystatin, and base medium with nystatin and deadcells) per box. Ten mL of live BY4741xBY4739, grown overnight in 1 mL of YPD, was used to inoculate120 wells on each of Day 0, 4, and 8 of the experiment, split evenly across boxes and medium types. OnDay 12, there was not enough live culture to inoculate all wells. Many control (YPD or SC) wells were notinoculated. One SCnystatin4 + dead cells well was not inoculated and excluded from the analysis. The otherwells were inoculated with as much as possible, ranging from 2 mL to 10 mL of culture (the fact that fewercells were used on Day 12 is conservative with respect to our results below). Between inoculation days, thestocks were stored at 4°C while the deep well boxes were incubated at 30°C, shaking at 200 rpm. The boxeswere visually inspected for growth every 24 hours for 16 days, excluding Days 10 and 11, using the samescoring system from zero to three as described in Section 4.2.5.4.3 ResultsIncluding all four acquisition experiments, we found that none of the 619 inoculated diploid wells (excludingthose found to have haploid contamination) or flasks underwent evolutionary rescue, compared with 116 outof 223MATa wells (52%) and 100 out of 308MAT↵ wells (32%) (Fig 4.2). These numbers are based on testsof resistance to nystatin performed by following the growth of the populations for 72 hours in their originalmedium type (either YPDnystatin4 or SCnystatin4) in the Bioscreen C. We conclude that ploidy playeda substantial role in the likelihood of evolutionary rescue. For each of the three acquisition experimentsin deep well boxes, strain type influenced the proportion of tested populations that were determined to beputatively resistant (2 contingency test, Acquisition Experiment 1: 2 = 53.14, df = 2, p-value = 3⇥ 1012;Acquisition Experiment 2: 2 = 119.36, df = 2, p-value < 1015; Acquisition Experiment 3: 2 = 193.98,df = 2, p-value < 1015). MATa consistently had the highest proportion of putative mutants among the wellsthat grew in the original acquisition experiments (1: 26/42 = 0.62; 2: 14/16 = 0.88, 3: 76/77 = 0.99), followedbyMAT↵ (1: 42/86 = 0.49; 2: 38/48 = 0.79, 3: 20/34 = 0.59). Only four of the 81 diploid populations testedfrom Acquisition Experiment 1 were deemed to be putative mutants, and none from Acquisition Experiments2 or 3 (Fig 4.2 and Fig C.2). These four potential mutant populations may be weakly resistant (allowing themto grow in lower concentrations of nystatin), but they were also determined to harbour haploid contaminantsat low frequencies, explaining the growth observed in YPDnystatin4. These results indicate that ploidyrestricts the ability of yeast populations to undergo evolutionary rescue under these conditions.The relative absence of adaptive evolution among diploids is unlikely to be due to a small number ofreplicates tested. We modelled the growth of a population from a single cell established on a plate to theportion of liquid that is used in the inocula for the acquisition experiments and expect 0.82 nucleotide-changing mutations to have occurred in at least one cell per site within the genome, combining all deep wellacquisition experiments and the flask experiment (Section C.2). When we multiply this over the averagelength of an ORF, we expect to have sampled ~888 different non-synonymous or nonsense mutations for754.3. Results●Diploid MATa MATαFigure 4.2: Plot indicating the number of potential mutants found in all wells over all three acquisitionexperiments in the deep well boxes. The area of the lightest coloured circle is proportional to the totalnumber of wells inoculated of each type. The intermediately coloured circle is proportional to the totalnumber of wells that showed some growth in the initial acquisition experiments. The darkest colouredcircle is proportional to the number of wells considered to carry potential mutants based on the follow-up growth assays. Very few of the diploid populations grew in the follow-up assays (green; 4/302/623;putative mutants/wells that grew/wells inoculated), while the majority of haploid populations that grew in theacquisition experiments were reliable mutants (red: MATa 116/135/223; blue: MAT↵ 100/168/308). Furthertesting indicated that these four diploid populations harboured haploid contamination at a low frequency.each gene within the genome over the course of the experiment. Based on these numbers, we believe thatif a one-step dominant rescue mutation were available in the diploids, we should have seen full resistanceevolve. If a secondary mutation is required in the same gene (either a second mutation in the same ORFor a loss of heterozygosity of the first mutation), there is a <0.0003 chance that such a secondary mutantcell would have been sampled over the course of the experiment (Section C.2). Thus, this experiment testsprimarily for single-step rescue mutations and not for the potential effects of homozygous mutations.4.3.1 Further testing of potential diploid mutantsFour diploid populations that grew in Acquisition Experiment 1 were considered to potentially carry rescuemutations (DR1-4 for “Diploid Rescue”). Two of these (DR1, 2) grew in all three replicate Bioscreen Cruns when testing to confirm resistance, while the other two (DR3, 4) grew in only two of the three follow-up assays. When colonies from these populations were retested under the original acquisition conditions(YPDnystatin4 in a deep well box), only a small amount of growth was observed for DR1-3 and no growthwas observed for DR4 in 72 hours. One colony from each of these populations (chosen among those thatgrew for DR1-3) was then assayed for growth in the Bioscreen, this time using freshly diluted nystatin, and nogrowth was observed for any of these strains in YPDnystatin4 while all had full growth in YPD. The nystatin764.4. Discussionused in the previous assays was older and possibly slightly degraded, potentially explaining the observedgrowth of DR1-4. These results indicate that the populations may harbour weakly resistant or higher fitnessmutants, but not mutations that are fully resistant to 4 mM nystatin. Further, follow-up investigation of thesepopulations from the end of the original growth assays indicated that low frequency haploid contaminationwas present in the original populations.4.3.2 Nystatin efficacy over timeBased on the observation that many populations grew in the initial acquisition experiments but were notresistant when re-tested in nystatin (see Section C.1), we hypothesized that nystatin was losing efficacy overthe course of the acquisition experiments (lasting seven to twelve days). Both YPDnystatin4 and SCnystatin4showed significant degradation of nystatin efficacy over time (Fig 4.3). This was observed as a loss of theability to inhibit the growth of BY diploids within four days. The presence or absence of dead yeast cellsin the medium did not have a significant effect on this loss of efficacy but the base medium did, with SCallowing more growth overall than YPD (generalized linear model with a binomial error distribution andlogit link function performed using the glm function in the package stats in R [R CORE TEAM 2016], p-values determined based on sequential likelihood ratio tests run using the anova function in the order: day ofinoculation: df = 3, p < 1015; base medium: df = 1, p = 0.0048; presence of dead cells: df = 1, p = 0.056).4.3.3 Rescue in larger populationsOne reason for a population to fail to undergo evolutionary rescue is that potentially adaptive mutationsare rare and do not occur in that population. To help determine whether mutational opportunity was alimiting factor preventing evolutionary rescue in diploids, we conducted a mutant acquisition experimentwith roughly 100-fold more initial cells (inoculation used 1 mL of overnight culture compared to 10 mL usedin the main box acquisition experiments). Ten diploid populations, as well as one population of each haploidtype, were exposed to 100 mL of YPDnystatin4 in 250 mL flasks. All populations grew within 10 days ofinoculation, with the MATa and MATa populations growing on Day 3, and the diploid populations growingon Days 8-10. Despite this growth, no diploid population grew when tested in YPDnystatin4 in the follow-up growth assay (Fig C.1). On the other hand, haploid populations generally showed reliable rescue asdemonstrated by growth in follow-up assays, consistent with the smaller volume experiments. These resultsindicate that increasing initial population size roughly 100-fold was not sufficient to allow for beneficialmutations to occur in diploids during this short adaptation experiment.4.4 DiscussionIn this study, we have provided an example where ploidy alters the probability of evolutionary rescue.Diploids were generally not able to adapt to a high concentration of nystatin in the short time course givenwith none of 619 inoculated wells (excluding those found to have haploid contamination) showing evolu-tionary rescue, as compared with 216 out of 531 haploid wells. Although we expected rescue to be less774.4. Discussion●● ●●●●● ●Day of inoculationPercentage grown in 4 days0 4 8 12020406080100●●SCSC+cellsYPDYPD+cellsFigure 4.3: Plot of the percentage of wells inoculated on each day that showed growth within four daysof inoculation in the nystatin efficacy experiment. Vertical bars represent 95% confidence intervals of theproportions. No wells grew within four days of the beginning of the experiment (when the nystatin wasfresh). When four day-old nystatin was inoculated with yeast, many wells grew within four days, and alleight-day old nystatin wells grew within four days. The increased variability in growth observed amongwells inoculated on Day 12 is likely due to the varying amounts of culture used to inoculate these wells(ranging from 2 mL to 10 mL). These results confirm that the nystatin may have been losing efficacy over thecourse of the initial acquisition experiments.784.4. Discussioncommon in diploids, given previous work demonstrating that mutations accumulated in haploids were reces-sive (GERSTEIN et al. 2014), it was possible that diploids would explore beneficial mutations not observedin the haploids. We conclude that there is a genetic limit to adaptation in diploids with no simple one-stepdominant rescue mutations available.There are many differences between haploids and diploids that may cause the observed difference in theirability to undergo evolutionary rescue. Factors that may favour haploid rescue include larger population size,differing effect sizes of mutations depending on ploidy background (if mutations tend to be more beneficialin haploids), and low dominance of potentially beneficial mutations (OTTO and WHITTON 2000). Thesedifferences will be discussed in more detail below.In yeast, haploids have smaller cells and therefore larger population sizes for the same volume of in-oculum (MABLE 2001), with a less than two-fold difference. Larger population sizes in haploids shouldcorrespond to a larger number of mutations in these populations, but this is not the case because diploidshave twice the genome (and therefore twice the number of mutational targets) when compared to haploids.To determine whether our starting population sizes were too small to allow diploid evolution, we exposedten diploid populations to nystatin in flasks containing 100-fold more medium and initial inoculum. None ofthese populations underwent evolutionary rescue (Fig C.1). At this population size (roughly 7⇥ 107 cells),we expect 38% of sites to mutate within at least one of the flasks, with over 600 mutations per gene expectedacross the flasks, leading us to conclude that initial population size is not the problem (Section C.2).The effect of potentially beneficial mutations may differ between haploids and diploids, even betweenhaploids and homozygous diploids, which are often treated as equivalent. Previous work in this system hasfound that the effect sizes of beneficial mutations were not equal between haploids and homozygous diploids,for those mutations acquired in a haploid background, with haploids outperforming diploids (GERSTEIN2013). In nystatin, these effect size differences may arise as a consequence of the geometrical differencesbetween haploid and diploid cells. Nystatin acts by binding to ergosterol in the yeast membrane, makingthe membrane more permeable to ions, sugars and metabolites, resulting in cell death (CARRILLO-MUÑOZet al. 2006). Because surface areas are higher for diploids than haploids (MABLE 2001), their sensitivityto nystatin may differ. In addition, the primary path to resistance to nystatin is through mutations in theergosterol biosynthesis pathway, which affects the sterol composition of the cell membrane (GERSTEIN et al.2012). Due to their larger surface area, diploids may suffer more of a fitness cost from these mutations dueto decreased stability of the membrane, thus impacting the effective fitness benefit conferred. Nevertheless,ergosterol mutations still confer a large fitness benefit to diploids in nystatin when present in homozygousform (GERSTEIN 2013), suggesting that the difference in rescue probability between haploids and diploidsis not due to differences in the effect of mutations when homozygous.Another way in which the effect of potentially beneficial mutations can differ between ploidies is throughtheir dominance. Many new mutations are recessive and therefore are only ‘seen’ by selection when rare ina haploid. Recessive beneficial alleles, especially those originating from new mutations, are unlikely torescue a diploid population because they will not often spread to high enough frequency for homozygotesto be common (ORR and UNCKLESS 2008). Sex by random assortment is very unlikely to combine rare,potentially beneficial recessive alleles. Natural yeast perform a version of selfing wherein mating is most794.4. Discussioncommon between gametes from a single diploid individual. This mechanism could produce an adapted,homozygous individual, but sex cannot be induced in our experimental setup, due to the short time frame andstrong selection. Because reproduction is strictly asexual in our experiment, diploids must acquire a secondmutation (either another new mutation or a loss-of-heterozygosity event) in order to gain any advantage froma recessive allele (MANDEGAR and OTTO 2007). Surprisingly, based on our results, we infer that there areno dominant or semi-dominant rescue mutations in this environment. This places a limit to evolutionaryrescue on diploids at a lower concentration of the drug than for haploids. These results will not necessarilygeneralize to other environments because they depend on the nature of available adaptive mutations and theexact effects of the environment on the ancestral type.ANDERSON et al. (2004) performed a similar experiment at high concentrations of another antifungaldrug (64 and 128 mg/mL fluconazole), in which the initial lines were able to undergo seven to nine doublingsbut were unable to proliferate further. In their experiment, diploids evolved resistance more slowly thanhaploids but eventually all replicate diploid lines evolved heritable resistance (minimum inhibitory concen-trations of 256 mg/mL fluconazole). Thus, diploids were not limited in their ability to undergo evolutionaryrescue under their experimental conditions. Importantly, we have similar total numbers of cells in our inocula(accounting for the initial growth that was observed in fluconazole), so that the contrast between observing100% rescue (ANDERSON et al. 2004) and our result of 0% rescue (for levels of drug used in the acquisitionphase) must reflect a difference in limits to evolutionary rescue and not a difference in experimental power.While certain two-step mutations would rescue diploid populations in our experiment (such as mutations inboth copies of an ergosterol pathway gene), they would require much larger populations than those used.We calculate a <0.0003 probability that a cell containing a secondary mutation in the same gene would besampled across any of the deep wells or flasks in the experiment, assuming a relatively high rate of eithersecondary mutation or loss of heterozygosity (~104) and assuming that any second mutation in the sameORF and/or mitotic recombination event would generate resistance, to be conservative (Section C.2). In con-trast, ANDERSON et al. (2004) observed patterns consistent with such two-step mutations. The difference inobserved two-step mutations likely reflects a difference in the selective environment. Because yeast undergoseveral generations in fluconazole before arresting growth, there is the opportunity for weakly resistant het-erozygous mutations to increase in frequency in the populations, and therefore a higher probability of lossof heterozygosity for one of these mutations.It may be the case that deterioration of the nystatin environment allows diploids to grow in our ex-periments sooner than they are able to adapt genetically. Phenotypic heterogeneity (without an underlyinggenetic basis) in the ability to persist in the presence of antibiotics has been observed in bacterial populations(e.g., BALABAN et al. 2004). These “persister” types remain sensitive to the antibiotic, however, when re-tested. Such persistence could explain the presence of nystatin-sensitive populations among those that grewin the initial acquisition experiments. The populations may persist at low numbers while the concentrationof nystatin is high enough to be inhibitory and then show growth once the efficacy of nystatin has droppedbelow some threshold. This is consistent with the observation that diploids tend to grow on the later daysof the acquisition experiments, which is when we also observe growth of non-resistant haploid populations(Fig. C.2). Follow up experiments evaluating the efficacy of nystatin over time indicate that nystatin likely804.4. Discussionlost efficacy by this time (Fig. 4.3), allowing the growth of lower tolerance strains. While we conclude thatfull resistance to 4 mM nystatin was not exhibited by any of the diploid lines assayed, it remains possible thatthe diploids did evolve low levels of resistance that improved their ability to persist or to grow once nystatinbecame less effective.In our study, we appear to have exceeded the limit of genetic adaptation possible in diploids, but not inhaploids, by using a high concentration of the fungicide nystatin and a short time frame. A previous studyfound that diploids were able to adapt at the same rate as haploids to a lower concentration of nystatin (0.6mM) over a longer period of time (140 generations) (GERSTEIN et al. 2011) under conditions that allowedgrowth of the initial strains (i.e., not an evolutionary rescue experiment). It is possible that the larger numberof generations in that experiment provided the opportunity for strains to get the kinds of two-step mutationsthat seem to be necessary for resistance to high concentrations of the drug. However, initial whole-genomesequence data from these strains found no mutations in either ERG3 or ERG6 (data not shown), the mostcommonly used genes in haploids at high concentrations of nystatin (GERSTEIN et al. 2012). Instead, theability of the diploids to evolve in GERSTEIN et al. (2011) suggests that different, and potentially moredominant, mutations may be available at lower concentrations of nystatin that are not sufficient to provideresistance to higher concentrations.We find that evolutionary rescue is not always possible and that the limits can depend strongly on theploidy of the organism in question. These results have implications for conservation. For example, amongalgae, we might expect evolutionary rescue in the face of climate change to change the relative proportions ofspecies with a haploid phase (haplonts or haploid-diploid species) relative to those with only a diploid phase(diplonts). As another example, higher standards may be needed for pollutants/toxins that require exposedorganisms to adapt using recessive mutations, because of the risk that evolutionary rescue will fail. Thereare also implications for disease management. By investigating the genetic basis of potential resistance toour treatments of choice (antibiotics, pesticides), we can make informed decisions about timing and dosage.In this way, we can endeavour to minimize potentially harmful evolutionary rescue in the organisms that weare attempting to control.81Chapter 5DiscussionIn this thesis, I have broadly addressed questions about the genetics of adaptation and speciation using theyeast Saccharomyces cerevisiae as a model system. I have used the tools of experimental design and statisticsalong with laboratory techniques for yeast manipulation and assessment to investigate the genetic basis ofevolution, taking advantage of the fast generation time of yeast and its genetic tractability. By allowing theorganism to explore possible mutations naturally and then sequencing to discover the utilized mutations, wecan begin to understand why evolution proceeded the way that it did. The use of experimental evolutiongreatly facilitated this process, as studies of natural systems are often limited by the precision with whichthey can map causative mutations, usually only able to narrow down the genome to large blocks of interest.For many species, these studies can often be restricted by either having relatively little knowledge availableabout the underlying genes or only having the ability to investigate candidate genes. I found that the geneticdetails of evolution are sometimes surprising, and they force us to expand our thinking about how evolution‘typically’ proceeds. Theoretical models of evolution must, by their nature, make certain assumptions aboutthe underlying genetic system, and I hope that results from these experiments will inform theoreticians aboutinteresting aspects of genetics that warrant further investigation. The chapters of my thesis have investigatedthe repeatability of adaptation in experimental strains of yeast adapted to high concentrations of copper(Chapter 2), tested genetic interactions between first-step adaptive mutations (Chapter 3), and explored thelimits of genetic adaptation of haploid and diploid yeast in a fungicide (Chapter 4). I will discuss the mainconclusions of these chapters in turn and then use Fisher’s geometric model as a theoretical framework torevisit some of the data from two of these chapters.5.1 Thesis summary5.1.1 Chapter 2: Repeatability of adaptationIn Chapter 2, I explore the repeatability of adaptation to high concentrations of copper. I find that the level ofrepeatability depends on what one defines as repeatable. The same gene (CUP1) was involved in increases incopy number in almost all replicates (27/34), as expected from previous work on copper resistance in yeast(ADAMO et al. 2012; FOGEL and WELCH 1982; FOGEL et al. 1983), so we might conclude that evolutionwas highly repeatable. The mechanism of copy number increase varied between strains, however, withsome strains utilizing whole-chromosome aneuploidy (an additional copy of the relevant chromosome) andothers increasing copy number of the already tandemly repeated region. Thus, at the level of mutation type,adaptation was slightly less repeatable. Finally, many strains also carried secondary genic mutations. Of theaffected genes, some were mutated in a few different strains (highly unlikely by chance alone), while some825.1. Thesis summarygenes were mutant in only one. These genes had a variety of functions and cellular localizations. Whenalso considering these genic mutations, the level of repeatability was quite low. These results highlight animportant contribution of experimental evolution to evolutionary theory. Observations like this force us toconsider what is meant by the term ‘repeated’, both in terms of the gene involved and the mutation type,highlighting the importance of the varying genetic mechanisms available to evolution.By comparing the results of Chapter 2 with those of GERSTEIN et al. (2012), which uses the same ex-perimental setup and same starting yeast strain but a different selective environment (a high concentration ofthe fungicide nystatin), we can observe how genomic breadth, and therefore repeatability, changes with theagent of selection. Few studies have directly compared the repeatability of adaptation between environmentsin this way (but see GRESHAM et al. 2008), but it is difficult to draw conclusions about the effects of a sin-gle factor (environment) when comparing studies that have been performed with many other varying factors(e.g., population size, population dynamics, organism of study). In our comparison, I find a variety of mu-tation types in copper (copy number variation, aneuploidy, genic mutations) while GERSTEIN et al. (2012)only find genic mutations in nystatin. Considering the biological pathways targetted, evolution in copper in-volved many different pathways and biological functions, while adaptation to nystatin was acquired throughmutations in genes of a single biosynthetic pathway. Finally, at the level of individual genes implicated inadaptation, increased copy number of a single gene, CUP1, was almost uniformly observed in evolution incopper. It is likely that amplification of this locus had a high mutation rate because it is already present in atandemly repeated region, which is prone to unequal crossover, gene conversion or single-strand annealing(ZHANG et al. 2013). Other genes were repeatedly involved in copper adaptation, but much less frequently.In contrast, adaptation to nystatin involved one of four genes, and most adapted strains carried mutations inone of the two repeatedly used genes (ERG3 or ERG6). Overall, adaptation to nystatin was more repeatablethan adaptation to copper at both the level of mutation type and biological pathway. At the level of individualgene, adaptation to copper was simultaneously more repeatable, due to the almost uniform observation ofCUP1 amplification, and less repeatable, due to the observation of a variety of other genic mutations involvedin multiple cellular processes.Unfortunately, due to the high mutability of the tandemly repeated CUP1 region, it was difficult todetermine the effects of genic mutations that co-occurred with high CUP1 copy numbers since lines withhigh CUP1 copy number were unstable (did not segregate 2:2 in cross progeny). For repeatedly hit genes,the benefit of these mutations was supported by many lines of evidence, including their presence in multipleindependently adapted strains, but it would have been much more difficult to characterize all of the othergenic mutations observed, and this was not done. Most studies that focus on genetic repeatability mainlyrefer to repeatably used genes. Going forward, it would be insightful to characterize the rarely-used genesin evolution to determine whether they are rare because the mutations have small fitness benefits or becausethe mutations themselves are rare, or whether it is often a combination of the two.5.1.2 Chapter 3: Evolution of BDM incompatibilities between first-step adaptive mutationsIn Chapter 3, I investigate epistasis between first-step adaptive mutations in the fungicide nystatin. Epistasisbetween beneficial mutations from the same environment can have implications for evolutionary trajectories,835.1. Thesis summaryaffecting both the speed and direction of evolutionary change, and is relevant for speciation. Independentlyevolving populations may adapt to similar selection pressures via different genetic changes, even if they areexposed to an identical environmental challenge. How these genetic changes interact in hybrids between thepopulations may determine whether or not the populations evolve to become new species. I used mutationsfrom an initial study of haploid adaptation to nystatin (GERSTEIN et al. 2012), focusing on one mutation ineach of four genes in the ergosterol biosynthesis pathway. I found that genetic interactions were prevalent andpredominantly negative, with the majority of mutations causing lower growth when combined in a doublemutant than when alone in a single mutant and, in one third of cases, the growth of the double mutant waslower than either single mutant. Thus, BDM incompatibilities evolved readily, even among populationsadapting to identical conditions. The prevalence of these kinds of interactions is surprising given the smallnumber of mutations tested and demonstrated that postzygotic reproductive isolation could evolve betweenpopulations differing by only a single genetic change each. These results lend support to the mutation-ordermodel of speciation where populations accumulate reproductive isolation due to the chance order in whichmutations arise in each. This model is difficult to prove in nature due to the requirement of known parallelselection histories and has not been thoroughly investigated in the lab. Further, the observation of isolatingepistasis between first-step mutations may drive the evolution of further divergence if one considers that theseearly mutations will potentially constrain future evolutionary paths to become increasingly incompatible.The observation of sign epistasis runs counter to expectations for mutations arising in a single biosyn-thetic pathway in the face of a simple selective pressure. We would expect these mutations to mask eachother (the upstream mutation masking the effects of the downstream one) (AVERY and WASSERMAN 1992).Similarly, if we use metabolic network theory to predict the epistasis between different genes in a linearpathway (SZATHMÁRY 1993), we would expect that if two genes independently reduce flux through a path-way then the double mutant should reduce flux more than either one alone but less than the combined effects(diminishing returns epistasis, not sign epistasis). The prevalence of BDM incompatibility-type interactionsindicates that the situation is not simply described by the linear biosynthetic pathway. Along these lines, wealso found that the phenotype of the double mutants was not reflected by the pathway position of the singlemutations, with sterol profiles of double mutants often matching one of the two single mutants but not alwaysthe upstream one. The sterol profile relationships did not match the fitness relationships either, implying thatgenes are deviating from the described pathway and that the mutations are changing something about thecell other than just the sterol profile in order to affect fitness.The nature of the genetic interactions depended not only on the mutations involved but also the concen-tration of drug in the assay conditions. When two strongly beneficial mutations were combined, I foundthat all double mutants had equivalent or lower fitness than the two parent single mutants in a non-stressfulor mildly-stressful environment, resulting in negative genetic interactions. When the stress (concentrationof nystatin) was increased, performance of the double mutants reversed; they were often the most fit strainand, at very high drug concentrations, were the only ones able to survive and grow, resulting in very positivegenetic interactions. This result is not initially intuitive but has theoretical grounding in Fisher’s geometricmodel of adaptation, as will be explored in Section 5.2.1, below. This was an especially interesting findingto make as I am not aware of many other similar examples, especially among BDM incompatibility-type845.1. Thesis summaryinteractors, and demonstrates the sensitivity of reproductive isolation to the environment in which hybridsare formed.The scope of this study was limited, however, and investigated only six pairwise combinations of alleles.In addition, these results may be specific to the nature of the mutations involved. The prevalence of negativeepistasis might come from the fact that all mutations investigated were in the ergosterol biosynthesis pathway.Negative epistasis can be observed as a result of two partial loss of function mutations in an essential pathwayif each decreases flux through the pathway (BOONE et al. 2007), which could explain our observationsbecause ergosterol is the main sterol in the yeast membrane and is therefore ostensibly essential. In orderto determine the generality of these conclusions, similar studies should be performed for a larger set ofbeneficial mutations across a variety of environmental conditions.5.1.3 Chapter 4: Limits to adaptationIn Chapter 4, I compare adaptation of haploids and diploids in the fungicide nystatin. I set out to investigatethe different genetic paths taken by haploid and diploid yeast evolving in a concentration of nystatin thatinhibits growth. As described briefly above, previous work in the lab (GERSTEIN et al. 2012) found thathaploid strains adapt to this environment by acquiring mutations in one of four genes of the ergosterolbiosynthesis pathway. To determine how diploid evolution might differ, we repeated the experiment usinghaploids of both mating types and diploids. There are many reasons why adaptation might proceed differentlyin haploid and diploid yeast including differences in number, distribution and types of mutations (due togenome size), dominance of adaptive mutations, fewer diploid cells in the same volume of culture due tocell size (MABLE 2001), and the same mutations having different effects or effect sizes in haploids vs.homozygous diploids (as found for some nystatin-resistance mutations by GERSTEIN 2013).Instead, I appear to have found the limit for diploid genetic adaptation through one-step mutations. Wefound no cases in which a diploid population evolved full resistance to the original evolutionary conditions.I do not believe that sampling effort was limiting in this case. We found no diploid mutants after testing over600 wells, compared with finding 216 haploid mutants out of 531 wells tested. In addition, I did not find adiploid mutant in any of the ten flasks tested, compared with both haploid flasks exhibiting genetic rescue. Itseems that diploid yeast are not able to obtain single mutations of large enough effect to allow growth. Thereis previous evidence that diploids are able to adapt at the same rate as haploids to a lower concentration ofnystatin (0.6 mM, which does not fully inhibit growth), over a longer time course (GERSTEIN et al. 2011),but the same is not true at this higher concentration (4 mM). All adaptive mutations previously found in thisinhibitory concentration of nystatin in haploids were recessive (GERSTEIN et al. 2014). Likely, the onlysingle-step mutations that have a large enough effect to genetically rescue a population in this environmentare recessive, and two mutational events at a single resistance-conferring gene are too rare to rescue thetested populations. This chapter provides an example of how the adaptive limit of an organism can depend onboth the exact environment in question (concentration of nystatin) and the biology of the organism (ploidy),having implications for the role of ploidy in evolutionary rescue in the face of a strong selective pressuresuch as an antibiotic.Interpretation of the results of this experiment was difficult because, in many instances, growth was855.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientsobserved without the evolution of heritable resistance to the original environmental conditions. In orderto isolate first-step, large-effect mutations, we used a concentration of nystatin that inhibited the growth ofwild-type yeast. In this environment, if yeast were to be serially transferred into fresh medium, they may bedriven to extinction by dilution before being given the opportunity to adapt. That is why we did not transferyeast over the course of the experiment but instead allowed them to remain in the original medium untilgrowth was observed. Unfortunately, because nystatin is not stable over long periods of time in 30°C, theefficacy of the drug decreased over the course of the experiment, leading to the growth of many non-resistantor partially-resistant yeast populations. Deciding on a cutoff for evolutionary rescue was therefore difficult,because some populations appear to be at their borderline for growth under the test conditions. In addition,we could not allow diploids more time to adapt because the environment was not stable for longer periodsof time. I would suggest that this experiment be repeated in other environments that are more likely to bestable for extended periods of time, or using careful addition of drugs over time, to test the generality ofthese conclusions.5.2 Putting adaptive mutations in the context of Fisher’s geometric modelover environmental gradientsIn this section, I would like to revisit some of the results from Chapter 2 and Chapter 3 in the context ofFisher’s geometric model (FISHER 1930). Fisher’s geometric model has been used to describe theoreticalreasons for epistasis between adaptive mutations (BLANQUART et al. 2014) and for patterns observed inspeciation (FRAÏSSE et al. 2016). It has also been combined with experimental work to describe patterns inthe mutations that arise in different concentrations of an antibiotic and how the optimum shifts between thoseconcentrations in Escherichia coli (HARMAND et al. 2017), as well as to test diminishing-returns epistasisin a multicellular fungus, Aspergillus nidulans (SCHOUSTRA et al. 2016). I will take a less formal approachthan the above-mentioned papers and use Fisher’s geometric model as a framework to think about patternsof epistasis and how the fitness of mutants might change in different levels of the same stressor.In Fisher’s geometric model, the fitness of a genotype is determined by its distance in phenotype spacefrom the optimal phenotype for that environment as well as a function that describes how fitness dropsoff with increasing distance from the optimum. Let’s imagine a population starting some distance fromthe fitness optimum in an environment. If it gains a beneficial mutation, we can imagine the mutation asa vector in n-dimensional phenotype space that brings the population closer to the fitness optimum. Thismutation fixes and now, when a second mutation arises, it is determined to be beneficial if it again bringsthe population closer to the fitness optimum. An important point to consider, however, is that this mutationwould not necessarily have been beneficial in the original genetic background, depending on the directionand size of the first mutation vector. For the same reason, two independently adaptive mutations (bothbeneficial in the original genetic background) may not be compatible with each other if, when combined,they cause the organism to overshoot the optimum (adding both vectors together results in a phenotypethat is further from the fitness optimum than either vector alone, Fig. 5.1). This is one example of howepistasis for fitness can arise in this model despite additivity of the underlying mutations and phenotypes,865.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientsFigure 5.1: Optimum overshooting (sign epistasis) in Fisher’s geometric model. Using Fisher’s geometricmodel of adaptation (FISHER 1930), we can imagine two genotypes (red and blue yeast) that are adaptingfrom a common ancestor (yellow yeast) to one optimum (the centre of the grey target). The initial beneficialmutations (red and blue arrows), represented as vectors in 2-dimensional space, put the genotypes in differentlocations in phenotype space. Later mutations will be beneficial or deleterious depending on the geneticbackground in which they arise and whether they bring the genotype closer to the fitness optimum. The twoinitial beneficial mutations are mutually exclusive, not conferring a fitness benefit in each other’s background,resulting in a less fit hybrid combination (purple yeast) that overshoots the phenotypic optimum.and we observe the outcome as sign epistasis in fitness between independently adaptive mutations. In thisway, evolutionary trajectories can be constrained by the size and direction of mutation vectors in phenotypespace, and reproductive isolation due to genetic incompatibility can arise between independently evolvingpopulations.I will attempt to describe two datasets using a simple form of Fisher’s geometric model. First, I willinvestigate the single and double mutant fitness data in different concentrations of nystatin from Chapter 3,to see whether the patterns of epistasis for fitness are consistent with underlying phenotypic additivity. Sec-ond, I will use the data measuring tolerance to copper from Chapter 2 to see whether fitness in a range ofconcentrations of copper can be predicted from a strain’s fitness in a single concentration of copper giveninformation about how other strains react across the range. Both of these analyses will utilize data from a setof genotypes measured in a series of concentrations of the stressor to which they have adapted. In Fisher’sgeometric model, changing environments are described by changing optima, and there are two main ways inwhich I will discuss an optimum changing (Fig. 5.2). The first is a shifting of the optimum as environmentschange. In this case, the optimum shifts in multidimensional phenotypic space, making different phenotypeseither beneficial or deleterious depending on their location in that space. The second is a narrowing (orwidening) of the optimum. In this case, the most fit phenotypes will remain the most fit, but relative andabsolute fitness values change.In order to see how well Fisher’s geometric model describes the observed data, we will need to place ourmutants in phenotype space. We will use fitness data for each strain, preferably in their adaptive environment,875.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientsFigure 5.2: Optima change with changing environments. In Fisher’s geometric model, different environ-ments are described by different positions and distributions of the phenotypic optimum. There are two mainways in which I might expect the optimum to change. First, as illustrated on the left, the optimum mightshift in location in different environmental conditions. Second, as illustrated on the right, the optimum mightnarrow (or widen) with changing conditions, making certain genotypes more or less fit without changingtheir distance to the optimum. In both cases, the initial fitness surface is represented by a light blue targetand the dark blue target is the fitness surface after an environmental change.to determine the relative positions of the strains. Imagine a one-dimensional line drawn between the ancestorand the fitness optimum. All first-step adaptive mutations will be vectors that originate from the ancestor,but they may point in any number of directions and cover any amount of distance, as long as they end closerto the optimum. However, the projection of those (potentially multi-dimensional) vectors onto the one-dimensional line drawn between the ancestor and the optimum will determine the fitness benefit conferredby each mutation (Fig. 5.3). I will therefore use differences in fitness measurements as proxies for distancefrom the ancestor towards the optimum in one-dimensional space. While it is possible to overshoot theoptimum (while still ending closer to the optimum than where the ancestor started), I will assume that thishas not occurred in the adaptive environment. This assumption is supported by the fact that the ancestorcannot grow in the selective environment and is therefore likely to be initially far from the optimum. I willalso assume linear mapping of fitness differences onto distance travelled in this one dimension, which doesnot have to be the case. I will use this technique to map mutants onto phenotype space. I will then keepphenotypic distances between strains constant as I investigate what occurs when the concentration of thestressor is changed.5.2.1 Can changing epistasis be explained by a shifting optimum? Revisiting Chapter 3In Chapter 3, we observed changing epistasis for optical density (OD) after 24 hours of growth with changingconcentrations of nystatin. When no nystatin was present in the medium, the ancestor was most fit but allstrains grew well. When low to moderate concentrations of nystatin were present, the single and doublemutants all had relatively high growth, indicating negative epistasis because double mutants were less fitthan expected. Finally, as the concentration of nystatin was increased, double mutants were often the onlystrains capable of growth. These observations led us to conclude that incompatibilities (as determined bysign epistasis) for growth rate that were present at moderate concentrations of nystatin (2 mM) were not stable885.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientsFigure 5.3: Projecting mutational effects into one dimension. If we were to draw a straight line connectingthe ancestral genotype (yellow yeast) and the fitness optimum (centre of the grey target), we could thenproject all of the mutational effects onto this dimension. This is shown for the red mutation (solid line). Byprojecting the red mutation onto the black line by way of the small dashed arrow, we can obtain the distancetravelled by that mutation in only the dimension of interest (red dashed arrow). The distance travelled alongthis one dimension is the relevant distance for fitness benefits conferred by the mutation.across different concentrations of the drug.I have replotted the OD data using the method described in the previous section in order to determinewhether these fitness measurements are consistent with underlying additivity at the phenotypic level anda shifting optimum with changing concentrations of the drug. First, to place the single mutants along thex-axis dimension, the difference in OD after 24 hours of growth between the single mutants and the ancestorwas found for 4 mM nystatin, the adaptive environment. Then, I placed the double mutant strains a distancefrom the ancestor that was the sum of the distances from the ancestor of the two single mutants of whichthe double mutant is composed. For example, the geneX geneY double mutant would be found at a distanceof 10 ‘units’ from the ancestor if geneX were 2 ‘units’ from the ancestor and geneY were 8 ‘units’ from theancestor. Once strains were positioned along the x-axis, I plotted the OD after 24 hours of growth for allconcentrations of nystatin (Fig. 5.4).When we look at the data, we notice that the method of placing strains along a single dimension worksrelatively well to produce interpretable results under Fisher’s geometric model. For each concentration ofnystatin, we can imagine a single fitness peak, and the fitness of the strains drops off roughly consistentlywith their distance from this peak along the x-axis. Because this remains true for each concentration ofnystatin without changing the positions of the strains along the x-axis, the data is broadly consistent with apeak changing in this one dimensional space. In contrast, we might have found that the strains needed tochange relative positions along the x-axis in order to find a single fitness peak, which would be interpretedas the relevant one-dimensional axis itself shifting in phenotype space (if the target in Fig. 5.3 were to moveoff of the black line, the projection of the red mutation onto the black line would no longer be useful forordering mutations relative to the optimum). There are two main implications of the consistency of the data895.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientsFigure 5.4: Changing epistasis in OD after 24 hours of growth in an environmental gradient. Mean OD after24 hours of growth in different concentrations of nystatin is plotted by strain. First-step adaptive mutationsin nystatin are plotted along the x-axis according to their proportional fitness benefit in 4 mM nystatin (theoriginal evolutionary environment, olive green) when compared with the ancestor (grey box). Double mutantstrains are plotted a distance from the ancestor equal to the sum of the distance travelled by each of theircomposite single mutations along the x-axis. Strains are represented by coloured boxes indicating theirmutations, with single mutants represented by a single coloured box (green: erg5, yellow: erg3, blue: erg7and red: erg6) and double mutants represented by two coloured boxes stacked together. Each concentrationof nystatin is plotted with a unique colour (see key on the right). On the bottom is a visual representationof how the optimum may be changing in different concentrations of nystatin. Only three optima are shown,for simplicity, and each is coloured according to the concentration that they represent. Both shifting andnarrowing are represented as the concentration of nystatin increases.905.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientswith Fisher’s geometric model. First, our results are broadly consistent with an underlying additive basisfor double mutant phenotypes. The phenotypic trait could arise from both single mutations decreasing fluxthrough the ergosterol biosynthesis pathway. We would expect two flux-reducing mutations to further reduceflux when combined as a double mutant, and this relationship could be additive on the one-dimensional phe-notype axis. Second, the changes in fitness of the different strains as the concentration of nystatin increasesseem to result from a combination of both shifting and narrowing of the fitness optimum. At very high con-centrations of nystatin (e.g., 64 mM), narrowing of the optimum is most obviously observed centred near theerg3 erg6 double mutant strain (yellow and red box in Fig. 5.4), which retains its ability to grow despite allother strains failing to do so. Shifting of the optimum is observed as only strains on the right-hand side ofthe plot maintaining their ability to grow in moderate to high concentrations of nystatin. If we again focus onthe erg3 erg6 double mutant, despite its high fitness at high concentrations of nystatin, there is no evidenceof this same strain being the most fit at low and moderate concentrations of nystatin, which is consistentwith an optimum that is shifting. The epistatic relationships between mutations arise as a consequence of thecombination of additivity on the phenotype axis (x-axis) and the position of the fitness optimum.From the plotted data, it seems likely that the sign epistasis observed for growth rate in 2 mM nystatinwas due to overshooting of the optimum in double mutant strains. This overshooting is difficult to observefrom the OD data in 2 mM nystatin, but growth rate is probably a more sensitive measure of fitness, whichallows better discrimination of fitness differences. Especially at lower concentrations of nystatin, wheremost strains are still capable of some growth, by waiting 24 hours until taking the OD measurement, we areleaving time for slower strains to ‘catch up’ in growth. Unfortunately, due to space constraints, we werenot able to perform growth rate analyses in a large number of concentrations of nystatin (the plotted data isunderlain by 3,840 measured wells, while only 400 wells can be simultaneously measured for growth rate).Similarly, the growth rate data presented in Chapter 3 includes more replicates per strain than this OD data,so we have greater confidence in each data point.5.2.2 Can the fitness of a strain in an environmental gradient be predicted from its fitnessin a single environment? Revisiting Chapter 2Inspired by the results of Section 5.2.1, I wanted to find out whether we could generally order adaptedstrains along a one-dimensional phenotypic axis by knowing their evolutionary environment and determinehow the optimum phenotype changes over an environmental gradient. If this were possible, then we couldtheoretically place newly discovered strains along that axis based on their fitness in one environment, andpotentially infer their relative fitness in the rest of the gradient. In order to determine whether this would beplausible, I revisited the dataset from Chapter 2 where copper tolerance was measured for all adapted ‘CopperBeneficial Mutation’ (CBM) lines. In this chapter, tolerance was calculated as the inhibitory concentration 50(IC50), which was determined by the OD of all strains after 72 hours of growth over a range of concentrationsof copper.I plotted the underlying OD data as described above with one important exception. I could not use theoriginal evolutionary environment (12.5 mM copper) to order the strains because many strains did not havehigher fitness than the ancestor in this environment. All strains have increased copper tolerance relative to915.2. Putting adaptive mutations in the context of Fisher’s geometric model over environmental gradientsOptical Density at 72hr●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●ancestor Copper Beneficial Mutation Lines0.00.51.01.50mM4mM8mM9mM10mM11mM12mM14mMFigure 5.5: OD of copper beneficial mutation lines after 72 hours of growth in a gradient of copper con-centrations. Mean OD measured after 72 hours of growth in different concentrations of copper is plottedby strain. CBM lines are plotted along the x-axis according to their proportional fitness benefit in 8 mMcopper (olive green) when compared with the ancestor (left hand side). Note that the original evolutionaryenvironment (12.5 mM copper) could not be used in this case because many strains did not have a higher ODthan the ancestor in this environment. 8 mM was chosen as the environment closest to the ancestor’s IC50.Each concentration of copper is represented by a unique colour (see key on the right).the ancestor (Fig. 2.2), but none have tolerance to 12.5 mM when measured after 72 hours. Their abilityto grow in the initial acquisition experiment under these conditions was likely due to a prolonged period ofsequestration of copper by tolerant cells, which might have decreased the effective concentration of copperin the medium and allowed for rapid proliferation. Indeed, the first copper-adapted lines were only isolatedseven days after inoculation, which is much later than the similarly acquired nystatin-adapted lines fromGERSTEIN et al. (2012) that were all isolated within seven days of incubation. Because I could not usethe data from 12.5 mM copper to order the strains, I instead used the data from 8 mM copper, as thisconcentration was the closest to the ancestor’s IC50 (Fig. 2.2). The results are plotted in Fig. 5.5.Based on Fig. 5.5, I would expect the predictive power of this dataset to be quite low. There is no clearsingle-peaked optimum at most concentrations of copper, and it does not shift or narrow in any broadlyobservable way. There are many reasons why this might be the case. First, as mentioned above, I couldnot use the adaptive environment to order the mutants. By using a lower concentration of copper, I mighthave chosen an environment where the actual optimum is somewhere intermediate compared to the truedistribution of the strains, thus breaking the assumption that we can determine the length of each mutationalong the single dimension by the difference in fitness between that strain and the ancestor. In fact, even ifwe did know the ‘true’ underlying fitness optimum in the original adaptive environment, we may have strainsthat have overshot the optimum (while still remaining fitter than the ancestor) in the original set. Many more925.3. Conclusionsmutations were observed per strain in copper than in nystatin, and the nature of the CUP1 mutation is suchthat it may actually change over time in the strain while we are measuring it (placing progeny of a single testat different positions along the axis). Additionally, our assumption that one phenotypic dimension shouldbe able to order mutants in relation to the optimum may not hold for this set of lines as there are manymore types of mutations (copy number variation in CUP1, chromosomal aneuploidy) in addition to genicmutations among these strains, and many more biological processes involved. The complexity observed maynecessitate the use of multiple dimensions to properly map the mutations in phenotype space.The nature of the selective environment may also be an important difference between the two datasetsexplored. While nystatin only has negative effects on the cells, copper does not. Cells require some amountof copper to survive. In fact, strains perform better in 4 mM copper than 0 mM (including the ancestor; notethat this is copper addition on top of what is normally present in rich YPD medium) and many strains areeven more fit in higher concentrations (compare Fig. 5.5 with Fig. 5.4 where growth is generally best in 0mM nystatin, and does not show large improvements in higher concentrations). The main mutation confer-ring copper tolerance is amplification of CUP1, which is a metallothionein protein that binds copper. It ispossible that having many copies of CUP1 in the cell is resulting in too many copper ions being sequesteredaway in lower concentrations of copper, but allowing for better utilization of copper in high concentrations(even better than in ‘non-stressful’ conditions). The complicated relationship between environmental con-centration and fitness is making fitness unpredictable in the one-dimensional case. We might have successby partitioning mutations along two axes, depending on whether or not they are in the CUP1 locus (similarto what was done in HARMAND et al. 2017 for gyrA and non-gyrA mutants). Finally, because OD was mea-sured after 72 hours of growth in this assay, it is possible that early fitness benefits/deficits in these strainsthat would be better described by Fisher’s geometric model are being masked by later growth. The lackof generality of this method of placing mutant strains in phenotype space that can be translated into fitnessby a singular phenotypic optimum indicates that there is still much to learn about the relationships betweengenotype, phenotype and fitness and how they are affected by evolution.5.3 ConclusionsMy work has implications for understanding of the genetic basis of adaptation in different types of en-vironments, levels of the same environment, and genetic backgrounds or ploidies. I find that the geneticrepeatability of adaptation depends on the genomic target size in the adaptive environment and that this tar-get size can also be influenced by the ploidy of the organism. I also investigate genetic interactions betweenadaptive mutations and find that different mutations adapting populations to the same environment mightoften lead to BDM-type incompatibilities, with consequences for the likelihood of mutation-order specia-tion. Finally, I find that the nature of those genetic interactions depends on the environment in which theyare measured, with negative interactions in the fungicide nystatin becoming positive in higher concentra-tions of the drug. The observed changes in interactions are consistent with a shifting and narrowing fitnessoptimum under Fisher’s geometric model in a one-dimensional phenotype space. When attempting to simi-larly place copper-adapted strains in a one-dimensional phenotype space, I found that simple fitness optima935.3. Conclusionswere not readily apparent. The usage of Fisher’s geometric model for predicting the changes in fitness ofadapted strains over environmental gradients may be limited by the biological nature of the interaction of theorganism with the environment in question, as mediated by the adaptive mutations.I believe that future investigations into the links between genotype, phenotype and fitness are importantfor progressing our models of evolution. Determining how mutations can lead to specific changes to pheno-type and fitness may depend on understanding these links at the molecular or biochemical level. We havemade great strides in our ability to broadly predict genes involved in adaptation (e.g., ergosterol pathwaygenes in nystatin resistance and CUP1 in copper resistance), but our ability to predict specific aspects ofadaptation (e.g., precise fitness effects of mutations, the type of epistasis between them, and their sensitivityto concentrations of a stressor) remains poor. By integrating molecular and biochemical information aboutadaptive mutations, including whether they act in biochemical pathways or protein complexes and their reg-ulatory relationships, we may improve our ability to map genotype onto phenotype. Generalities in theseprinciples would inform theoretical models of evolution, such as Fisher’s geometric model. As our ability todetermine the genetic basis of evolution grows with the increasing ease of genomic studies, the next step isto improve our mechanistic understanding of how those genetic changes lead to changes in fitness.94BibliographyABRAMOFF, M., P. MAGALHAES, and S. RAM, 2004 Image processing with ImageJ. Biophotonics Inter-national 11: 36–42.ADAMO, G. M., S. BROCCA, S. PASSOLUNGHI, B. SALVATO, and M. LOTTI, 2012 Laboratory evolutionof copper tolerant yeast strains. Microbial Cell Factories 11: 1.ADAMS, J., and F. ROSENZWEIG, 2014 Experimental microbial evolution: history and conceptual under-pinnings. Genomics 104: 393–398.AKEN, B. L., P. ACHUTHAN, W. AKANNI, M. R. AMODE, F. BERNSDORFF, et al., 2017 Ensembl 2017.Nucleic Acids Research 45: D635–D642.ALEXANDER, H. K., G. MARTIN, O. Y. MARTIN, and S. BONHOEFFER, 2014 Evolutionary rescue: linkingtheory for conservation and medicine. Evolutionary Applications 7: 1161–1179.ALEXANDER, R. P., G. FANG, J. ROZOWSKY, M. SNYDER, and M. B. GERSTEIN, 2010 Annotatingnon-coding regions of the genome. Nature Reviews. Genetics 11: 559–571.ALONSO, J. M., A. N. STEPANOVA, T. J. LEISSE, C. J. KIM, H. CHEN, et al., 2003 Genome-wide inser-tional mutagenesis of Arabidopsis thaliana. Science 301: 653–657.ANDERSON, J., J. FUNT, D. THOMPSON, S. PRABHU, A. SOCHA, et al., 2010 Determinants of divergentadaptation and Dobzhansky-Muller interaction in experimental yeast populations. Current Biology 20:1383–1388.ANDERSON, J. B., C. SIRJUSINGH, and N. RICKER, 2004 Haploidy, diploidy and evolution of antifungaldrug resistance in Saccharomyces cerevisiae. Genetics 168: 1915–23.ANGERT, A. L., H. BRADSHAW JR, and D. W. SCHEMSKE, 2008 Using experimental evolution to investi-gate geographic range limits in monkeyflowers. Evolution 62: 2660–2675.ARNEGARD, M. E., M. D. MCGEE, B. MATTHEWS, K. B. MARCHINKO, G. L. CONTE, et al., 2014Genetics of ecological divergence during speciation. Nature 511: 307–311.ARTHINGTON-SKAGGS, B. A., H. JRADI, T. DESAI, and C. J. MORRISON, 1999 Quantitation of ergosterolcontent: novel method for determination of fluconazole susceptibility of Candida albicans. Journal ofClinical Microbiology 37: 3332–3337.95BibliographyAVERY, L., and S. WASSERMAN, 1992 Ordering gene function: the interpretation of epistasis in regulatoryhierarchies. Trends in Genetics 8: 312–316.BABA, T., T. ARA, M. HASEGAWA, Y. TAKAI, Y. OKUMURA, et al., 2006 Construction of Escherichiacoli K-12 in-frame, single-gene knockout mutants: the Keio collection. Molecular Systems Biology 2:2006.0008.BAILEY, S. F., and T. BATAILLON, 2016 Can the experimental evolution programme help us elucidate thegenetic basis of adaptation in nature? Molecular Ecology 25: 203–218.BALABAN, N. Q., J. MERRIN, R. CHAIT, L. KOWALIK, and S. LEIBLER, 2004 Bacterial persistence as aphenotypic switch. Science 305: 1622–1625.BALAKRISHNAN, R., J. PARK, K. KARRA, B. C. HITZ, G. BINKLEY, et al., 2012 YeastMine–an integrateddata warehouse for Saccharomyces cerevisiae data as a multipurpose tool-kit. Database 2012: bar062.BANDYOPADHYAY, S., M. MEHTA, D. KUO, M.-K. SUNG, R. CHUANG, et al., 2010 Rewiring of geneticnetworks in response to DNA damage. Science 330: 1385–1389.BARKER, B., L. XU, and Z. GU, 2015 Dynamic epistasis under varying environmental perturbations. PloSOne 10: e0114911.BARRICK, J. E., and R. E. LENSKI, 2013 Genome dynamics during experimental evolution. Nature Re-views. Genetics 14: 827.BARTON, N., and L. PARTRIDGE, 2000 Limits to natural selection. BioEssays 22: 1075–1084.BATES, D., M. MÄCHLER, B. BOLKER, and S. WALKER, 2015 Fitting linear mixed-effects models usinglme4. Journal of Statistical Software 67: 1–48.BATESON, W., 1909 Mendel’s principles of heredity. Cambridge University Press.BEADLE, G. W., and E. L. TATUM, 1941 Genetic control of biochemical reactions in Neurospora. Proceed-ings of the National Academy of Sciences 27: 499–506.BELL, G., 2013 Evolutionary rescue and the limits of adaptation. Philosophical Transactions of the RoyalSociety B 368: 20120080.BELL, G., and S. COLLINS, 2008 Adaptation, extinction and global change. Evolutionary Applications 1:3–16.BELL, G., and A. GONZALEZ, 2009 Evolutionary rescue can prevent extinction following environmentalchange. Ecology Letters 12: 942–948.BELLEN, H. J., R. W. LEVIS, Y. HE, J. W. CARLSON, M. EVANS-HOLM, et al., 2011 TheDrosophila genedisruption project: progress using transposons with distinctive site specificities. Genetics 188: 731–743.96BibliographyBEN-ARI, G., D. ZENVIRTH, A. SHERMAN, L. DAVID, M. KLUTSTEIN, et al., 2006 Four linked genesparticipate in controlling sporulation efficiency in budding yeast. PLoS Genetics 2: e195.BERNARDES, J. P., R. B. STELKENS, and D. GREIG, 2017 Heterosis in hybrids within and between yeastspecies. Journal of Evolutionary Biology 30: 538–548.BIKARD, D., D. PATEL, C. LE METTÉ, V. GIORGI, C. CAMILLERI, et al., 2009 Divergent evolution ofduplicate genes leads to genetic incompatibilities within A. thaliana. Science 323: 623–626.BLANQUART, F., G. ACHAZ, T. BATAILLON, and O. TENAILLON, 2014 Properties of selected mutationsand genotypic landscapes under Fisher’s geometric model. Evolution 68: 3537–3554.BLOWS, M. W., and A. A. HOFFMANN, 2005 A reassessment of genetic limits to evolutionary change.Ecology 86: 1371–1384.BOONE, C., H. BUSSEY, and B. J. ANDREWS, 2007 Exploring genetic interactions and networks withyeast. Nature Reviews. Genetics 8: 437–449.BORDENSTEIN, S. R., and M. D. DRAPEAU, 2001 Genotype-by-environment interaction and the Dobzhan-sky–Muller model of postzygotic isolation. Journal of Evolutionary Biology 14: 490–501.BOTSTEIN, D., and G. R. FINK, 2011 Yeast: An experimental organism for 21st century biology. Genetics189: 695–704.BRESLOW, D. K., D. M. CAMERON, S. R. COLLINS, M. SCHULDINER, J. STEWART-ORNSTEIN, et al.,2008 A comprehensive strategy enabling high-resolution functional analysis of the yeast genome. NatureMethods 5: 711–718.BURCH, C. L., and L. CHAO, 1999 Evolution by small steps and rugged landscapes in the RNA virus phi6.Genetics 151: 921–927.BURKE, M. K., J. P. DUNHAM, P. SHAHRESTANI, K. R. THORNTON, M. R. ROSE, et al., 2010 Genome-wide analysis of a long-term evolution experiment with Drosophila. Nature 467: 587.BUSKIRK, S. W., R. E. PEACE, and G. I. LANG, 2017 Hitchhiking and epistasis give rise to cohort dynamicsin adapting populations. Proceedings of the National Academy of Sciences 114: 8330–8335.BYRNE, A. B., M. T. WEIRAUCH, V. WONG, M. KOEVA, S. J. DIXON, et al., 2007 A global analysis ofgenetic interactions in Caenorhabditis elegans. Journal of Biology 6: 8.C. ELEGANS DELETION MUTANT CONSORTIUM AND OTHERS, 2012 Large-scale screening for targetedknockouts in the Caenorhabditis elegans genome. G3: Genes, Genomes, Genetics 2: 1415–1425.CARLSON, S. M., C. J. CUNNINGHAM, and P. A. H. WESTLEY, 2014 Evolutionary rescue in a changingworld. Trends in Ecology & Evolution 29: 521–530.97BibliographyCARRILLO-MUÑOZ, A. J., G. GIUSIANO, P. A. EZKURRA, and G. QUINDÓS, 2006 Antifungal agents:mode of action in yeast cells. Revista Española de Quimioterapia 19: 130–139.CASPETA, L., Y. CHEN, P. GHIACI, A. FEIZI, S. BUSKOV, et al., 2014 Altered sterol composition rendersyeast thermotolerant. Science 346: 75–78.CHATR-ARYAMONTRI, A., R. OUGHTRED, L. BOUCHER, J. RUST, C. CHANG, et al., 2017 The BioGRIDinteraction database: 2017 update. Nucleic Acids Research 45: D369–D379.CHERRY, J. M., E. L. HONG, C. AMUNDSEN, R. BALAKRISHNAN, G. BINKLEY, et al., 2011 Saccha-romyces Genome Database: the genomics resource of budding yeast. Nucleic Acids Research 40: D700–D705.CHOU, H. H., J. BERTHET, and C. J. MARX, 2009 Fast growth increases the selective advantage of amutation arising recurrently during evolution under metal limitation. PLoS Genetics 5: e1000652.CHOU, H.-H., H.-C. CHIU, N. F. DELANEY, D. SEGRÈ, and C. J. MARX, 2011 Diminishing returnsepistasis among beneficial mutations decelerates adaptation. Science 332: 1190–1192.CHOU, H.-H., N. F. DELANEY, J. A. DRAGHI, and C. J. MARX, 2014 Mapping the fitness landscape ofgene expression uncovers the cause of antagonism and sign epistasis between adaptive mutations. PLoSGenetics 10: e1004149–11.CONRAD, T. M., A. R. JOYCE, M. K. APPLEBEE, C. L. BARRETT, B. XIE, et al., 2009 Whole-genomeresequencing of Escherichia coli K-12 MG1655 undergoing short-term laboratory evolution in lactateminimal media reveals flexible selection of adaptive mutations. Genome Biology 10: R118.CONTE, G. L., M. E. ARNEGARD, J. BEST, Y. F. CHAN, F. C. JONES, et al., 2015 Extent of QTL reuseduring repeated phenotypic divergence of sympatric threespine stickleback. Genetics 201: 1189–1200.CONTE, G. L., M. E. ARNEGARD, C. L. PEICHEL, and D. SCHLUTER, 2012 The probability of geneticparallelism and convergence in natural populations. Proceedings of the Royal Society B 279: 5039–5047.COSTANZO, M., B. VANDERSLUIS, E. N. KOCH, A. BARYSHNIKOVA, C. PONS, et al., 2016 A globalgenetic interaction network maps a wiring diagram of cellular function. Science 353: aaf1420–aaf1420.COVO, S., C. M. PUCCIA, J. L. ARGUESO, D. A. GORDENIN, and M. A. RESNICK, 2014 The sister chro-matid cohesion pathway suppresses multiple chromosome gain and chromosome amplification. Genetics196: 373–384.COYNE, J., and H. ORR, 2004 Speciation. Sinauer Associates, Incorporated Publishers.CREGG, J. M., 2007 DNA-mediated transformation. Pichia Protocols : 27–42.CUTTER, A., 2005 Mutation and the experimental evolution of outcrossing in Caenorhabditis elegans. Jour-nal of Evolutionary Biology 18: 27–34.98BibliographyDARWIN, C., 1859 On the origin of species by means of natural selection, or, The preservation of favouredraces in the struggle for life. London: John Murray.DARWIN, C., and A. WALLACE, 1858 On the tendency of species to form varieties; and on the perpetuationof varieties and species by natural means of selection. Zoological Journal of the Linnean Society 3: 45–62.DE VISSER, J. A. G. M., T. F. COOPER, and S. F. ELENA, 2011 The causes of epistasis. Proceedings ofthe Royal Society B 278: 3617–3624.DE VISSER, J. A. G. M., and D. E. ROZEN, 2005 Limits to adaptation in asexual populations. Journal ofEvolutionary Biology 18: 779–788.DETTMAN, J. R., J. B. ANDERSON, and L. M. KOHN, 2008 Divergent adaptation promotes reproductiveisolation among experimental populations of the filamentous fungus Neurospora. BMC EvolutionaryBiology 8: 35.DETTMAN, J. R., C. SIRJUSINGH, L. M. KOHN, and J. B. ANDERSON, 2007 Incipient speciation bydivergent adaptation and antagonistic epistasis in yeast. Nature 447: 585–588.DEUTSCHBAUER, A. M., and R. W. DAVIS, 2005 Quantitative trait loci mapped to single-nucleotide reso-lution in yeast. Nature Genetics 37: 1333–1340.DHAR, R., R. SÄGESSER, C. WEIKERT, J. YUAN, and A. WAGNER, 2011 Adaptation of Saccharomycescerevisiae to saline stress through laboratory evolution. Journal of Evolutionary Biology 24: 1135–1153.DIXON, S. J., M. COSTANZO, A. BARYSHNIKOVA, B. ANDREWS, and C. BOONE, 2009 Systematic map-ping of genetic interaction networks. Annual Review of Genetics 43: 601–625.DIXON, S. J., Y. FEDYSHYN, J. L. Y. KOH, T. S. K. PRASAD, C. CHAHWAN, et al., 2008 Significantconservation of synthetic lethal genetic interaction networks between distantly related eukaryotes. Pro-ceedings of the National Academy of Sciences 105: 16653–16658.DOWELL, R. D., O. RYAN, A. JANSEN, D. CHEUNG, S. AGARWALA, et al., 2010 Genotype to phenotype:a complex problem. Science 328: 469–469.EHRENREICH, I. M., J. BLOOM, N. TORABI, X. WANG, Y. JIA, et al., 2012 Genetic architecture of highlycomplex chemical resistance traits across four yeast strains. PLoS Genetics 8: e1002570.ELROD, S. L., S. M. CHEN, K. SCHWARTZ, and E. O. SHUSTER, 2009 Optimizing sporulation conditionsfor different Saccharomyces cerevisiae strain backgrounds. Methods in Molecular Biology 557: 21–26.ENTIAN, K.-D., T. SCHUSTER, J. HEGEMANN, D. BECHER, H. FELDMANN, et al., 1999 Functionalanalysis of 150 deletion mutants in Saccharomyces cerevisiae by a systematic approach. Molecular andGeneral Genetics 262: 683–702.99BibliographyFERNANDES, A. R., M. PRIETO, and I. SÁ-CORREIA, 2000 Modification of plasma membrane lipid orderand H+-ATPase activity as part of the response of Saccharomyces cerevisiae to cultivation under mild andhigh copper stress. Archives of Microbiology 173: 262–268.FERNANDES, A. R., and I. SÁ-CORREIA, 2001 The activity of plasma membrane H(+)-ATPase is stronglystimulated during Saccharomyces cerevisiae adaptation to growth under high copper stress, accompanyingintracellular acidification. Yeast 18: 511–521.FISHER, R. A., 1930 The Genetical Theory of Natural Selection. Oxford University Press, Oxford.FOGEL, S., and J. W. WELCH, 1982 Tandem gene amplification mediates copper resistance in yeast. Pro-ceedings of the National Academy of Sciences 79: 5342–5346.FOGEL, S., J. W. WELCH, G. CATHALA, and M. KARIN, 1983 Gene amplification in yeast: CUP1 copynumber regulates copper resistance. Current Genetics 7: 347–355.FOGLE, C. A., J. L. NAGLE, and M. M. DESAI, 2008 Clonal interference, multiple mutations and adapta-tion in large asexual populations. Genetics 180: 2163–2173.FRAÏSSE, C., P. A. GUNNARSSON, D. ROZE, N. BIERNE, and J. J. WELCH, 2016 The genetics of specia-tion: Insights from Fisher’s geometric model. Evolution 70: 1450–1464.FROST, A., M. G. ELGORT, O. BRANDMAN, C. IVES, S. R. COLLINS, et al., 2012 Functional repurposingrevealed by comparing S. pombe and S. cerevisiae genetic interactions. Cell 149: 1339–1352.GARLAND, T., and M. R. ROSE, 2009 Experimental evolution: concepts, methods, and applications ofselection experiments. University of California Press.GERSTEIN, A. C., 2013 Mutational effects depend on ploidy level: all else is not equal. Biology Letters 9:20120614–20120614.GERSTEIN, A. C., L. A. CLEATHERO, M. A. MANDEGAR, and S. P. OTTO, 2011 Haploids adapt fasterthan diploids across a range of environments. Journal of Evolutionary Biology 24: 531–540.GERSTEIN, A. C., A. KUZMIN, and S. P. OTTO, 2014 Loss-of-heterozygosity facilitates passage throughHaldane’s sieve for Saccharomyces cerevisiae undergoing adaptation. Nature Communications 5: 3819.GERSTEIN, A. C., D. S. LO, and S. P. OTTO, 2012 Parallel genetic changes and nonparallel gene-environment interactions characterize the evolution of drug resistance in yeast. Genetics 192: 241–252.GIAEVER, G., A. M. CHU, L. NI, C. CONNELLY, L. RILES, et al., 2002 Functional profiling of theSaccharomyces cerevisiae genome. Nature 418: 387–391.GOMPEL, N., and B. PRUD’HOMME, 2009 The causes of repeated genetic evolution. Developmental Biol-ogy 332: 36–47.100BibliographyGOMPEL, N., B. PRUD’HOMME, P. J. WITTKOPP, V. A. KASSNER, and S. B. CARROLL, 2005 Chancecaught on the wing: cis-regulatory evolution and the origin of pigment patterns in Drosophila. Nature433: 481–487.GOMULKIEWICZ, R., and R. D. HOLT, 1995 When does evolution by natural selection prevent extinction?Evolution 49: 201–207.GOULD, S., 1989 Wonderful Life: The Burgess Shale and the Nature of History. Norton, New York.GRADEN, J. A., and D. R. WINGE, 1997 Copper-mediated repression of the activation domain in the yeastMac1p transcription factor. Proceedings of the National Academy of Sciences 94: 5550–5555.GRAMATES, L. S., S. J. MARYGOLD, G. D. SANTOS, J.-M. URBANO, G. ANTONAZZO, et al., 2017Flybase at 25: looking to the future. Nucleic Acids Research 45: D663–D671.GRESHAM, D., M. M. DESAI, C. M. TUCKER, H. T. JENQ, D. A. PAI, et al., 2008 The repertoire anddynamics of evolutionary adaptations to controlled nutrient-limited environments in yeast. PLoS Genetics4: e1000303.GROSS, B. L., and L. H. RIESEBERG, 2005 The ecological genetics of homoploid hybrid speciation. Journalof Heredity 96: 241–252.HARMAND, N., R. GALLET, R. JABBOUR-ZAHAB, G. MARTIN, and T. LENORMAND, 2017 Fisher’sgeometrical model and the mutational patterns of antibiotic resistance across dose gradients. Evolution71: 23–37.HARRISON, R., B. PAPP, C. PAL, S. G. OLIVER, and D. DELNERI, 2007 Plasticity of genetic interactionsin metabolic networks of yeast. Proceedings of the National Academy of Sciences 104: 2307–2312.HARROW, J., A. FRANKISH, J. M. GONZALEZ, E. TAPANARI, M. DIEKHANS, et al., 2012 Gencode: thereference human genome annotation for the encode project. Genome Research 22: 1760–1774.HARTMAN, J. L., B. GARVIK, and L. HARTWELL, 2001 Principles for the buffering of genetic variation.Science 291: 1001–1004.HERRON, M. D., and M. DOEBELI, 2013 Parallel evolutionary dynamics of adaptive diversification inEscherichia coli. PLoS Biology 11: e1001490.HICKMAN, M. A., G. ZENG, A. FORCHE, M. P. HIRAKAWA, D. ABBEY, et al., 2013 The ‘obligate diploid’Candida albicans forms mating-competent haploids. Nature 494: 55–59.HO, C. H., L. MAGTANONG, S. L. BARKER, D. GRESHAM, S. NISHIMURA, et al., 2009 A molecularbarcoded yeast orf library enables mode-of-action analysis of bioactive compounds. Nature Biotechnology27: 369–377.HOU, J., A. FRIEDRICH, J.-S. GOUNOT, and J. SCHACHERER, 2015 Comprehensive survey of condition-specific reproductive isolation reveals genetic incompatibility in yeast. Nature Communications 6: 7214.101BibliographyHUALA, E., A. W. DICKERMAN, M. GARCIA-HERNANDEZ, D. WEEMS, L. REISER, et al., 2001 TheArabidopsis information resource (TAIR): a comprehensive database and web-based information retrieval,analysis, and visualization system for a model plant. Nucleic Acids Research 29: 102–105.HUSE, H. K., T. KWON, J. E. A. ZLOSNIK, D. P. SPEERT, E. M. MARCOTTE, et al., 2010 Parallelevolution in Pseudomonas aeruginosa over 39,000 generations in vivo. mBio 1: e00199–10.HVORECNY, K. L., and G. PRELICH, 2010 A systematic cen library of the Saccharomyces cerevisiaegenome. Yeast 27: 861–865.JASNOS, L., K. TOMALA, D. PACZESNIAK, and R. KORONA, 2008 Interactions between stressful environ-ment and gene deletions alleviate the expected average loss of fitness in yeast. Genetics 178: 2105–2111.JAYAKODY, L. N., N. HAYASHI, and H. KITAGAKI, 2011 Identification of glycolaldehyde as the key in-hibitor of bioethanol fermentation by yeast and genome-wide analysis of its toxicity. BiotechnologyLetters 33: 285–292.JELIER, R., J. I. SEMPLE, R. GARCIA-VERDUGO, and B. LEHNER, 2011 Predicting phenotypic variationin yeast from individual genome sequences. Nature Genetics 43: 1270–1274.JENSEN-PERGAKES, K. L., M. A. KENNEDY, N. D. LEES, R. BARBUCH, C. KOEGEL, et al., 1998 Se-quencing, disruption, and characterization of the Candida albicans sterol methyltransferase (ERG6) gene:drug susceptibility studies in erg6 mutants. Antimicrobial Agents and Chemotherapy 42: 1160–1167.JONES, G. M., J. STALKER, S. HUMPHRAY, A. WEST, T. COX, et al., 2008 A systematic library forcomprehensive overexpression screens in Saccharomyces cerevisiae. Nature Methods 5: 239–241.JOSHI, C. J., and A. PRASAD, 2014 Epistatic interactions among metabolic genes depend upon environ-mental conditions. Molecular Biosystems 10: 2578–2589.KACHROO, A. H., J. M. LAURENT, A. AKHMETOV, M. SZILAGYI-JONES, C. D. MCWHITE, et al.,2017 Systematic bacterialization of yeast genes identifies a near-universally swappable pathway. eLife 6:e25093.KAO, K. C., and G. SHERLOCK, 2008 Molecular characterization of clonal interference during adaptiveevolution in asexual populations of Saccharomyces cerevisiae. Nature Genetics 40: 1499–1504.KASSEN, R., 2014 Experimental evolution and the nature of biodiversity. Roberts.KAWECKI, T. J., R. E. LENSKI, D. EBERT, B. HOLLIS, I. OLIVIERI, et al., 2012 Experimental evolution.Trends in Ecology & Evolution 27: 547–560.KESELER, I. M., A. MACKIE, A. SANTOS-ZAVALETA, R. BILLINGTON, C. BONAVIDES-MARTÍNEZ,et al., 2017 The EcoCyc database: reflecting new knowledge about Escherichia coli K-12. Nucleic AcidsResearch 45: D543–D550.102BibliographyKHAN, A. I., D. M. DINH, D. SCHNEIDER, R. E. LENSKI, and T. F. COOPER, 2011 Negative epistasisbetween beneficial mutations in an evolving bacterial population. Science 332: 1193–1196.KINSELLA, R. J., A. KÄHÄRI, S. HAIDER, J. ZAMORA, G. PROCTOR, et al., 2011 Ensembl BioMarts: ahub for data retrieval across taxonomic space. Database 2011: bar030–bar030.KISHONY, R., and S. LEIBLER, 2003 Environmental stresses can alleviate the average deleterious effect ofmutations. Journal of Biology 2: 14.KOMSTA, L., 2011 Outliers: Tests for outliers. R package version 0.14.KONO, K., A. AL-ZAIN, L. SCHROEDER, M. NAKANISHI, and A. E. IKUI, 2016 Plasma membrane/cellwall perturbation activates a novel cell cycle checkpoint during G1 in Saccharomyces cerevisiae. Proceed-ings of the National Academy of Sciences 113: 6910–6915.KRYAZHIMSKIY, S., D. RICE, E. JERISON, and M. DESAI, 2014 Global epistasis makes adaptation pre-dictable despite sequence-level stochasticity. Science 344: 1519–1522.KUMAR, A., N. BELOGLAZOVA, C. BUNDALOVIC-TORMA, S. PHANSE, V. DEINEKO, et al., 2016 Con-ditional epistatic interaction maps reveal global functional rewiring of genome integrity pathways in Es-cherichia coli. Cell Reports 14: 648–661.KVITEK, D. J., and G. SHERLOCK, 2011 Reciprocal sign epistasis between frequently experimentallyevolved adaptive mutations causes a rugged fitness landscape. PLoS Genetics 7: e1002056.LACHAPELLE, J., and G. BELL, 2012 Evolutionary rescue of sexual and asexual populations in a deterio-rating environment. Evolution 66: 3508–3518.LANDRY, C. R., J. P. TOWNSEND, D. L. HARTL, and D. CAVALIERI, 2006 Ecological and evolutionarygenomics of Saccharomyces cerevisiae. Molecular Ecology 15: 575–591.LANG, G. I., D. P. RICE, M. J. HICKMAN, E. SODERGREN, G. M. WEINSTOCK, et al., 2013 Pervasivegenetic hitchhiking and clonal interference in forty evolving yeast populations. Nature 500: 571–574.LARKIN, M. A., G. BLACKSHIELDS, N. BROWN, R. CHENNA, P. A. MCGETTIGAN, et al., 2007 ClustalW and Clustal X version 2.0. Bioinformatics 23: 2947–2948.LEE, A. Y., R. P. ST ONGE, M. J. PROCTOR, I. M. WALLACE, A. H. NILE, et al., 2014 Mapping thecellular response to small molecules using chemogenomic fitness signatures. Science 344: 208–211.LEE, I., Z. LI, and E. M. MARCOTTE, 2007 An improved, bias-reduced probabilistic functional genenetwork of baker’s yeast, Saccharomyces cerevisiae. PloS One 2: e988.LEES, N. D., B. SKAGGS, D. R. KIRSCH, and M. BARD, 1995 Cloning of the late genes in the ergosterolbiosynthetic pathway of Saccharomyces cerevisiae – a review. Lipids 30: 221–226.103BibliographyLEHNER, B., C. CROMBIE, J. TISCHLER, A. FORTUNATO, and A. G. FRASER, 2006 Systematic map-ping of genetic interactions in Caenorhabditis elegans identifies common modifiers of diverse signalingpathways. Nature Genetics 38: 896–903.LENSKI, R. E., 2017 Experimental evolution and the dynamics of adaptation and genome evolution inmicrobial populations. The ISME Journal 11: 2181–2194.LENTH, R. V., 2016 Least-squares means: The R package lsmeans. Journal of Statistical Software 69: 1–33.LEXER, C., Z. LAI, and L. H. RIESEBERG, 2004 Candidate gene polymorphisms associated with salttolerance in wild sunflower hybrids: implications for the origin of Helianthus paradoxus, a diploid hybridspecies. New Phytologist 161: 225–233.LI, H., B. HANDSAKER, A. WYSOKER, T. FENNELL, J. RUAN, et al., 2009 The sequence alignment/mapformat and SAMtools. Bioinformatics 25: 2078–2079.LINDSEY, H. A., J. GALLIE, S. TAYLOR, and B. KERR, 2013 Evolutionary rescue from extinction iscontingent on a lower rate of environmental change. Nature 494: 463–467.LITI, G., 2015 The natural history of model organisms: The fascinating and secret wild life of the buddingyeast S. cerevisiae. Elife 4: e05835.LOBKOVSKY, A. E., and E. V. KOONIN, 2012 Replaying the tape of life: quantification of the predictabilityof evolution. Frontiers in Genetics 3: 246.LONG, A., G. LITI, A. LUPTAK, and O. TENAILLON, 2015 Elucidating the molecular architecture ofadaptation via evolve and resequence experiments. Nature Reviews. Genetics 16: 567.LOSOS, J. B., 1992 The evolution of convergent structure in Caribbean Anolis communities. SystemsBiology 41: 403–420.LOURENÇO, M., R. S. RAMIRO, D. GÜLERESI, J. BARROSO-BATISTA, K. B. XAVIER, et al., 2016 Amutational hotspot and strong selection contribute to the order of mutations selected for during Escherichiacoli adaptation to the gut. PLoS Genetics 12: e1006420.LYNCH, M., W. SUNG, K. MORRIS, N. COFFEY, and C. LANDRY, 2008 A genome-wide view of thespectrum of spontaneous mutations in yeast. Proceedings of the National Academy of Sciences 105:9272–9277.MABLE, B., 2001 Ploidy evolution in the yeast Saccharomyces cerevisiae: a test of the nutrient limitationhypothesis. Journal of Evolutionary Biology 14: 157–170.MACLEAN, R. C., A. R. HALL, G. G. PERRON, and A. BUCKLING, 2010 The population genetics ofantibiotic resistance: integrating molecular mechanisms and treatment contexts. Nature Reviews. Genetics11: 405–414.104BibliographyMAHESHWARI, S., and D. A. BARBASH, 2011 The genetics of hybrid incompatibilities. Annual Review ofGenetics 45: 331–355.MALLET, J., 2007 Hybrid speciation. Nature 446: 279–283.MANCEAU, M., V. S. DOMINGUES, C. R. LINNEN, E. B. ROSENBLUM, and H. E. HOEKSTRA, 2010 Con-vergence in pigmentation at multiple levels: mutations, genes and function. Philosophical Transactions ofthe Royal Society B 365: 2439–2450.MANDEGAR, M., and S. OTTO, 2007 Mitotic recombination counteracts the benefits of genetic segregation.Proceedings of the Royal Society B 274: 1301–1307.MARTIN, A., and V. ORGOGOZO, 2013 The loci of repeated evolution: A catalog of genetic hotspots ofphenotypic variation. Evolution 67: 1235–1250.MATSUO, T., S. SUGAYA, J. YASUKAWA, T. AIGAKI, and Y. FUYAMA, 2007 Odorant-binding proteinsobp57d and obp57e affect taste perception and host-plant preference in Drosophila sechellia. PLoS Biol-ogy 5: e118.MAURICIO, R., 1998 Costs of resistance to natural enemies in field populations of the annual plant Ara-bidopsis thaliana. The American Naturalist 151: 20–28.MCCOURT, P., H.-Y. LIU, J. E. PARKER, C. GALLO-EBERT, M. DONIGAN, et al., 2016 Proper steroldistribution is required for Candida albicans hyphal formation and virulence. G3: Genes, Genomes,Genetics 6: 3455–3465.MCDONALD, M. J., D. P. RICE, and M. M. DESAI, 2016 Sex speeds adaptation by altering the dynamicsof molecular evolution. Nature 531: 233–236.MCDOWALL, M. D., M. A. HARRIS, A. LOCK, K. RUTHERFORD, D. M. STAINES, et al., 2014 Pombase2015: updates to the fission yeast database. Nucleic Acids Research 43: D656–D661.MERZ, S., and B. WESTERMANN, 2009 Genome-wide deletion mutant analysis reveals genes required forrespiratory growth, mitochondrial genome maintenance and mitochondrial protein synthesis in Saccha-romyces cerevisiae. Genome Biology 10: R95.MEYER, J. R., D. T. DOBIAS, J. S. WEITZ, J. E. BARRICK, R. T. QUICK, et al., 2012 Repeatability andcontingency in the evolution of a key innovation in phage lambda. Science 335: 428–432.MILLER, C., P. JOYCE, and H. WICHMAN, 2011 Mutational effects and population dynamics during viraladaptation challenge current models. Genetics 187: 185.MIRA, P. M., J. C. MEZA, A. NANDIPATI, and M. BARLOW, 2015 Adaptive landscapes of resistance geneschange as antibiotic concentrations change. Molecular Biology and Evolution 32: 2707–2715.MORTIMER, R. K., and J. R. JOHNSTON, 1986 Genealogy of principal strains of the yeast genetic stockcenter. Genetics 113: 35–43.105BibliographyMUKHOPADHYAY, K., A. KOHLI, and R. PRASAD, 2002 Drug susceptibilities of yeast cells are affected bymembrane lipid composition. Antimicrobial Agents and Chemotherapy 46: 3695–3705.MUSSO, G., M. COSTANZO, M. HUANGFU, A. M. SMITH, J. PAW, et al., 2008 The extensive andcondition-dependent nature of epistasis among whole-genome duplicates in yeast. Genome Research18: 1092–1099.NOSIL, P., and D. SCHLUTER, 2011 The genes underlying the process of speciation. Trends in Ecology &Evolution 26: 160–167.ONO, J., A. C. GERSTEIN, and S. P. OTTO, 2016 Data from: Widespread genetic incompatibilities betweenfirst-step mutations during parallel adaptation of Saccharomyces cerevisiae to a common environment.Dryad Digital Repository. Openly available via http://dx.doi.org/10.5061/dryad.vs370.ONO, J., A. C. GERSTEIN, and S. P. OTTO, 2017 Widespread genetic incompatibilities between first-step mutations during parallel adaptation of Saccharomyces cerevisiae to a common environment. PLoSBiology 15: e1002591.ORR, H. A., 1995 The population genetics of speciation: the evolution of hybrid incompatibilities. Genetics139: 1805–1813.ORR, H. A., and A. J. BETANCOURT, 2001 Haldane’s sieve and adaptation from the standing geneticvariation. Genetics 157: 875–884.ORR, H. A., and S. P. OTTO, 1994 Does diploidy increase the rate of adaptation? Genetics 136: 1475–1480.ORR, H. A., and R. L. UNCKLESS, 2008 Population extinction and the genetics of adaptation. The AmericanNaturalist 172: 160–169.OTTO, S. P., and J. WHITTON, 2000 Polyploid incidence and evolution. Annual Review of Genetics 34:401–437.PAPP, B., R. A. NOTEBAART, and C. PAL, 2011 Systems-biology approaches for predicting genomic evo-lution. Nature Reviews. Genetics 12: 591–602.PAYEN, C., A. B. SUNSHINE, G. T. ONG, J. L. POGACHAR, W. ZHAO, et al., 2016 High-throughput identi-fication of adaptive mutations in experimentally evolved yeast populations. PLoS Genetics 12: e1006339.PEÑA, M., J. LEE, and D. THIELE, 1999 A delicate balance: homeostatic control of copper uptake anddistribution. Journal of Nutrition 129: 1251–1260.POELWIJK, F. J., D. J. KIVIET, D. M. WEINREICH, and S. J. TANS, 2007 Empirical fitness landscapesreveal accessible evolutionary paths. Nature 445: 383.POELWIJK, F. J., S. TA˘NASE-NICOLA, D. J. KIVIET, and S. J. TANS, 2011 Reciprocal sign epistasis is anecessary condition for multi-peaked fitness landscapes. Journal of Theoretical Biology 272: 141–144.106BibliographyPRÁ, D., S. I. R. FRANKE, R. GIULIAN, M. L. YONEAMA, J. F. DIAS, et al., 2008 Genotoxicity andmutagenicity of iron and copper in mice. Biometals 21: 289–297.PRESGRAVES, D. C., 2010 The molecular evolutionary basis of species formation. Nature Reviews. Genet-ics 11: 175.PRUD’HOMME, B., N. GOMPEL, A. ROKAS, V. A. KASSNER, T. M. WILLIAMS, et al., 2006 Repeatedmorphological evolution through cis-regulatory changes in a pleiotropic gene. Nature 440: 1050.R CORE TEAM, 2015 R: A language and environment for statistical computing. R Foundation for StatisticalComputing, Vienna, Austria.R CORE TEAM, 2016 R: A language and environment for statistical computing. R Foundation for StatisticalComputing, Vienna, Austria.REMOLD, S. K., and R. E. LENSKI, 2004 Pervasive joint influence of epistasis and plasticity on mutationaleffects in Escherichia coli. Nature Genetics 36: 423–426.REPLANSKY, T., V. KOUFOPANOU, D. GREIG, and G. BELL, 2008 Saccharomyces sensu stricto as a modelsystem for evolution and ecology. Trends in Ecology & Evolution 23: 494–501.ROBINSON, M. D., J. GRIGULL, N. MOHAMMAD, and T. R. HUGHES, 2002 FunSpec: a web-based clusterinterpreter for yeast. BMC Bioinformatics 3: 35.ROGUEV, A., S. BANDYOPADHYAY, M. ZOFALL, K. ZHANG, T. FISCHER, et al., 2008 Conservation andrewiring of functional modules revealed by an epistasis map in fission yeast. Science 322: 405–410.ROTH, F. P., H. D. LIPSHITZ, and B. J. ANDREWS, 2009 Q&A: Epistasis. Journal of Biology 8: 35.ROZEN, D., J. DE VISSER, and P. GERRISH, 2002 Fitness effects of fixed beneficial mutations in microbialpopulations. Current Biology 12: 1040–1045.RYAN, C. J., P. CIMERMANCˇICˇ, Z. A. SZPIECH, A. SALI, R. D. HERNANDEZ, et al., 2013 High-resolutionnetwork biology: connecting sequence with function. Nature Reviews. Genetics 14: 865–879.RYAN, C. J., A. ROGUEV, K. PATRICK, J. XU, H. JAHARI, et al., 2012 Hierarchical modularity and theevolution of genetic interactomes across species. Molecular Cell 46: 691–704.SAMBROOK, J., and D. W. RUSSELL, 2001Molecular Cloning: A laboratory manual,. 3rd ed. Cold SpringHarbor Laboratory Press, New York, USA.SANGLARD, D., F. ISCHER, T. PARKINSON, D. FALCONER, and J. BILLE, 2003 Candida albicans mu-tations in the ergosterol biosynthetic pathway and resistance to several antifungal agents. AntimicrobialAgents and Chemotherapy 47: 2404–2412.SCHENK, M., I. SZENDRO, M. SALVERDA, J. KING, and J. DE VISSER, 2013 Patterns of epistasis betweenbeneficial mutations in an antibiotic resistance gene. Molecular Biology and Evolution 30: 1779–1787.107BibliographySCHLUTER, D., 2009 Evidence for ecological speciation and its alternative. Science 323: 737–741.SCHLUTER, D., E. A. CLIFFORD, M. NEMETHY, and J. S. MCKINNON, 2004 Parallel evolution andinheritance of quantitative traits. The American Naturalist 163: 809–822.SCHNEEBERGER, K., 2014 Using next-generation sequencing to isolate mutant genes from forward geneticscreens. Nature Reviews. Genetics 15: 662–676.SCHOUSTRA, S., S. HWANG, J. KRUG, and J. A. G. M. DE VISSER, 2016 Diminishing-returns epistasisamong random beneficial mutations in a multicellular fungus. Proceedings of the Royal Society B 283:20161376.SELMECKI, A. M., K. DULMAGE, L. E. COWEN, J. B. ANDERSON, and J. BERMAN, 2009 Acquisition ofaneuploidy provides increased fitness during the evolution of antifungal drug resistance. PLoS Genetics5: e1000705.SHERMAN, F., 2002 Getting started with yeast. Methods in Enzymology 350: 3–41.SIONOV, E., H. LEE, Y. C. CHANG, and K. J. KWON-CHUNG, 2010 Cryptococcus neoformans overcomesstress of azole drugs by formation of disomy in specific multiple chromosomes. PLoS Pathogens 6:e1000848.SMITH, J. M., R. BURIAN, S. KAUFFMAN, P. ALBERCH, J. CAMPBELL, et al., 1985 Developmental con-straints and evolution: A perspective from the mountain lake conference on development and evolution.The Quarterly Review of Biology 60: 265–287.SOARES, E. V., 2011 Flocculation in Saccharomyces cerevisiae: a review. Journal of Applied Microbiology110: 1–18.ST ONGE, R. P., R. MANI, J. OH, M. PROCTOR, E. FUNG, et al., 2007 Systematic pathway analysis usinghigh-resolution fitness profiling of combinatorial gene deletions. Nature Genetics 39: 199–206.STEIN, L., P. STERNBERG, R. DURBIN, J. THIERRY-MIEG, and J. SPIETH, 2001 Wormbase: networkaccess to the genome and biology of Caenorhabditis elegans. Nucleic Acids Research 29: 82–86.STELKENS, R. B., M. A. BROCKHURST, G. D. D. HURST, and D. GREIG, 2014 Hybridization facilitatesevolutionary rescue. Evolutionary Applications 7: 1209–1217.STERN, D. L., 2000 Perspective: evolutionary developmental biology and the problem of variation. Evolu-tion 54: 1079–1091.STERN, D. L., 2013 The genetic causes of convergent evolution. Nature Reviews. Genetics 14: 751–764.STROOBANTS, A., J.-M. DELROISSE, F. DELVIGNE, J. DELVA, D. PORTETELLE, et al., 2008 Isolationand biomass production of a Saccharomyces cerevisiae strain binding copper and zinc ions. AppliedBiochemistry and Biotechnology 157: 85–97.108BibliographySZATHMÁRY, E., 1993 Do deleterious mutations act synergistically? Metabolic control theory provides apartial answer. Genetics 133: 127–132.TENAILLON, O., A. RODRÍGUEZ-VERDUGO, R. L. GAUT, P. MCDONALD, A. F. BENNETT, et al., 2012The molecular diversity of adaptive convergence. Science 335: 457–461.TESTE, M.-A., M. DUQUENNE, J. M. FRANCOIS, and J.-L. PARROU, 2009 Validation of reference genesfor quantitative expression analysis by real-time RT-PCR in Saccharomyces cerevisiae. BMC MolecularBiology 10: 99.THOMPSON, O., M. EDGLEY, P. STRASBOURGER, S. FLIBOTTE, B. EWING, et al., 2013 The millionmutation project: a new approach to genetics in Caenorhabditis elegans. Genome Research 23: 1749–1762.TISCHLER, J., B. LEHNER, and A. G. FRASER, 2008 Evolutionary plasticity of genetic interaction net-works. Nature Genetics 40: 390–391.TKESHELASHVILI, L., T. MCBRIDE, K. SPENCE, and L. LOEB, 1991 Mutation spectrum of copper-induced DNA damage. Journal of Biological Chemistry 266: 6401–6406.TONG, Z., M.-S. KIM, A. PANDEY, and P. J. ESPENSHADE, 2014 Identification of candidate substrates forthe Golgi Tul1 E3 ligase using quantitative diGly proteomics in yeast. Molecular and Cellular Proteomics13: 2871–2882.WANG, A. D., N. P. SHARP, C. C. SPENCER, K. TEDMAN-AUCOIN, and A. F. AGRAWAL, 2009 Selec-tion, epistasis, and parent-of-origin effects on deleterious mutations across environments in Drosophilamelanogaster. The American Naturalist 174: 863–874.WEINREICH, D., R. WATSON, and L. CHAO, 2005 Perspective: sign epistasis and genetic constraint onevolutionary trajectories. Evolution 59: 1165–1174.WEIRAUCH, M. T., A. YANG, M. ALBU, A. G. COTE, A. MONTENEGRO-MONTERO, et al., 2014 Deter-mination and inference of eukaryotic transcription factor sequence specificity. Cell 158: 1431–1443.WELCH, J., D. MALONEY, and S. FOGEL, 1991 Gene conversions within the Cup1r region from heterolo-gous crosses in Saccharomyces cerevisiae. Molecular and General Genetics 229: 261–266.WENGER, J. W., J. PIOTROWSKI, S. NAGARAJAN, K. CHIOTTI, G. SHERLOCK, et al., 2011 Hunger artists:yeast adapted to carbon limitation show trade-offs under carbon sufficiency. PLoS Genetics 7: e1002202.WHITLOCK, M. C., P. C. PHILLIPS, F. B.-G. MOORE, and S. J. TONSOR, 1995 Multiple fitness peaks andepistasis. Annual Review of Ecology and Systematics 26: 601–629.WONG, A., N. RODRIGUE, and R. KASSEN, 2012 Genomics of adaptation during experimental evolutionof the opportunistic pathogen Pseudomonas aeruginosa. PLoS Genetics 8: e1002928.109BibliographyWOODS, R., 1971 Nystatin-resistant mutants of yeast: alterations in sterol content. Journal of Bacteriology108: 69–73.YAMPOLSKY, L. Y., and A. STOLTZFUS, 2005 The exchangeability of amino acids in proteins. Genetics170: 1459–1472.YONA, A. H., Y. S. MANOR, R. H. HERBST, G. H. ROMANO, A. MITCHELL, et al., 2012 Chromosomalduplication is a transient evolutionary solution to stress. Proceedings of the National Academy of Sciences109: 21010–21015.ZEYL, C., T. VANDERFORD, and M. CARTER, 2003 An evolutionary advantage of haploidy in large yeastpopulations. Science 299: 555–8.ZHANG, H., A. F. ZEIDLER, W. SONG, C. M. PUCCIA, E. MALC, et al., 2013 Gene copy-number variationin haploid and diploid strains of the yeast Saccharomyces cerevisiae. Genetics 193: 785–801.ZHAO, Y., P. K. STROPE, S. G. KOZMIN, J. H. MCCUSKER, F. S. DIETRICH, et al., 2014 Structuresof naturally evolved CUP1 tandem arrays in yeast indicate that these arrays are generated by unequalnonhomologous recombination. G3: Genes, Genomes, Genetics 4: 2259–2269.ZHU, Y. O., G. SHERLOCK, and D. A. PETROV, 2016 Whole genome analysis of 132 clinical Saccha-romyces cerevisiae strains reveals extensive ploidy variation. G3: Genes, Genomes, Genetics 6: 2421–2434.ZINSER, E., F. PALTAUF, and G. DAUM, 1993 Sterol composition of yeast organelle membranes and sub-cellular distribution of enzymes involved in sterol metabolism. Journal of Bacteriology 175: 2853–2858.110AppendixAppendix AAppendix for Chapter 2: Too much of agood thing: The unique and repeated pathstoward copper adaptationA.1 Supplementary MethodsA.1.1 Quantitative real-time PCR (qPCR)To test whether levels of CUP1 inferred from in silico qPCR were consistent with levels of CUP1 transcrip-tion, we assayed RNA levels using quantitative real-time PCR (qPCR). We chose 10 CBM lines that spannedthe range of CUP1 copy number (from lowest to highest): CBM16, CBM22, CBM24, CBM37, CBM2,CBM14, CBM51, CBM4, CBM34, and CBM13. For each line and BY4741, culture was struck from frozenonto a YPD plate and grown at 30C for 48h. A single colony of each CBM line and two colonies of BY4741were inoculated into 1mL YPD + 5.5mM copper (a lower concentration was used to allow growth of all lines,including BY4741) and grown for 12 hours at 30C with shaking, at which point RNA was isolated usingthe RNEasy Mini Kit from Qiagen, following the yeast protocol. cDNA was reverse transcribed from 500ngof RNA using MultiScribe reverse transcriptase (Life Technologies) and oligo d(T) primers.Oligonucleotides for qPCR (Table 2.2) were designed using Primer Express (ABI). mRNA levels ofTAF10 were used for normalization because TAF10 has stable expression across strains and conditions(TESTE et al. 2009). cDNA was diluted 100-fold for CUP1, but not TAF10, to account for differencesin their abundance in the samples. All qPCR reactions were performed using an ABI7000 sequence detec-tion system (Applied Biosystems, Inc.). The reaction volume was 22µL containing 10pmol of each primerand 2µL of each sample with 10µL of 2X SYBR Green Master Mix (Applied Biosystems Inc.). Reactionconditions were 1 cycle of 50C for 2 min., 1 cycle of 94C for 10 min., and 40 cycles of 95C for 15 sec.,60C for 1 min.To generate standard curves against which cDNA concentrations could be measured, we obtained cDNAfrom BY4741, which was then diluted five times, each time using a five-fold dilution, followed by qPCR us-ing the primers for TAF10 and CUP1. Standard curves for BY4741 were plotted such that a 1:1 relationship111A.1. Supplementary Methodsbetween fluorescence and sample concentration yields an expected slope of log2(10) = 3.32. All standardcurves fit the data well (r2 >0.99), with slopes between 3.1 and 3.3. TAF10 and CUP1 expression levelsfor every other strain were then measured against their respective standard curves, and then CUP1 levelswere divided by TAF10 levels to control for variation among samples in total cDNA concentrations. AllqPCR experiments were performed with two technical replicates.A.1.2 Tetrad analysesTo separate the effects of single mutations from other mutations present in the evolved lines (including extracopies of CUP1), we crossed all of the CBM lines with BY4739 (MAT↵ leu240 lys240 ura340), whichhas a common genotype yet opposite mating type and different auxotrophies than BY4741, the progenitorof our lines. Cells of both mating types were allowed time to mate overnight on a YPD plate before beingstruck onto plates lacking histidine and lysine, selecting for diploids. Single colonies were then struck ontoselection plates a second time to ensure they were diploids. Culture was taken directly from these plates andfrozen in 15% glycerol.To isolate single mutations, we attempted to sporulate the CBM⇥BY4739 lines that contained eachcommon mutation or aneuploidy and the fewest number of additional mutations (⇠1/3 of the lines). Weencountered substantial difficulties in obtaining tetrads from our strains; BY4741, a derivative of S288c, isknown to be a poor sporulator (BEN-ARI et al. 2006; DEUTSCHBAUER and DAVIS 2005). With a subsetof the lines, we attempted to maximize sporulation rates using a variety of different protocols including allcombinations of YPD, 1% YPA or 6% YEPD for pre-sporulation, liquid or plates, and PSP2, 1% KAc, orCSHSPO as the sporulation medium, all in liquid (see ELROD et al. 2009 for media details). Most combina-tions were tried at 30C, but the CSHSPO combinations were also attempted at 25C and 37C. In all cases,frozen culture was struck onto YPD plates and grown for 48 hours at 30C to isolate a single colony forsporulation. Ultimately, we obtained the most success using YPD liquid as a pre-sporulation medium fol-lowed by washing 100µL of overnight culture with dH2O and then plating on 1%KAc at 20C or 25C for upto 30 days. Unfortunately, however, we remained unable to sporulate the majority of lines. In particular, de-spite many attempts, no tetrads were obtained for CBM16 (PMA1 mutation plus chrII aneuploidy), CBM26(PMA1 mutation plus chrI, chrV and chrVIII aneuploidy), CBM29 (PMA1 mutation plus chrII aneuploidy),CBM47 (VTC1 mutation), or CBM55 (no mutation identified other than extra copies of CUP1).We were able to sporule CBM2 (chrII aneuploidy), CBM14 (MAM3 mutation), CBM25 (MLP1 andENA5 mutations), and CBM34 (VTC4 mutation). CBM25 was not initially chosen for tetrad dissection butwas dissected as a contaminate of CBM22 (VTC1 plus chrVIII and chrXVI aneuploidy), as detected bysubsequent sequencing. CBM25 contaminating cells were likely positively selected during the sporulationprocedure given that the aneuploid lines in our experiment, like CBM22, had very low sporulation rates. Theresulting tetrads were dissected by micromanipulation on YPD plates. The spores were allowed to germinateand grow at 30C for 3 days before each dissection plate was replica plated to test for mating types andauxotrophies. All tetrads were verified for 2:2 segregation of auxotrophies (except the aneuploid CBM2 -see below) and mating type. Once confirmed, the colonies obtained from each spore were frozen in 15%112A.1. Supplementary Methodsglycerol. We numbered the dissected tetrads (t1 up to t11) and lettered each haploid spore (a-d).The genotype of resulting spores was then determined. For CBM14 tetrad lines, MAM3 was amplifiedby PCR (primers in Table 2.2), and the product was digested with EcoRV (Fermentas), which specificallycuts the mutant allele at GATATC. The results (one band versus two) were visualized on a 2% agarosegel. CBM25 spores were sequenced on Illumina HiSeq 2000, which is when the strain was discovered tobe CBM25 (bearing a mutation in MLP1 and ENA5), not CBM22. For CBM34 spores, VTC4 was PCRamplified (Table 2.2) from genomic DNA and the fragment was Sanger sequenced using the forward primerand aligned to the reference sequence using ClustalW at EMBL-EBI (LARKIN et al. 2007). All SNPs showedthe expected 2:2 segregation pattern.The segregation pattern for the additional copy of chrII in CBM2 spores was determined for three of thetetrads (t1, t2 and t5) by the presence of the LYS2 alleles, as detected by PCR. The LYS2 gene is locatedon chrII, and the mated diploid from CBM2 carried two functioning copies of the gene (from CBM2) andone copy of the lys240 allele (from BY4739). Primers were designed to flank the LYS2 gene. The forwardprimer was designed 538bp upstream of the start site and the reverse primer was designed 487bp downstreamof the stop codon in order to easily detect the deletion by band size (full gene = 5199bp, deletion allele =708bp) (primers in Table 2.2). For t1 and t2, the two functional copies were inferred to be in the same celldue to the 2:2 segregation of the deletion and wild type alleles. For t5, one functional allele had segregatedto each cell and the two aneuploid cells were determined based on PCR detection of the presence of thedeletion. Corresponding phenotypes were verified by plating on medium lacking lysine. The segregationpattern for the additional copy of chrII in t3 was determined by Illumina sequencing, followed by calculatingthe total depth of coverage for each chromosome, as described above.Southern blots with CUP1 specific probes were performed to quantify the segregation patterns of CUP1among the spores. DNA concentration of genomic DNA isolated from each analyzed spore was measured intriplicate with the Qubit fluorometer (Invitrogen). Based on the average concentration, 2µg of each samplewas loaded into a 1% agarose gel and run at 120V. DNA in the gel was denatured in a NaOH buffer andtransferred to a nylon membrane (Hybond N+, GE Healthcare) using capillary transfer in 20x SSC buffer,affixing the DNA to the membrane by baking at 80C for 2 hours. The membrane was incubated overnight at57C, with two biotin labelled probes (Table 2.2). The membrane was then washed in 2x SSC + 0.1% SDSbuffer at 56C 3 times for 15min. Probe binding was visualized using the North2South chemiluminescentdetection kit (Thermo Scientific). Blots were exposed onto CL-XPosure Film (Thermo Scientific) for 30sec.to 1min. and developed in a Kodak X-ray film processor. We isolated genomic DNA and ran a Southern bloton three separate occasions for each spore. Controls (BY4741, BY4739, original CBM line) were always runin duplicate on the same gel as related spores. Band intensity was quantified in ImageJ (ABRAMOFF et al.2004) using the “background corrected density" macro (http://rsb.info.nih.gov/ij/macros/BackgroundCorrectedDensity.txt).113A.2. Supporting TablesA.2 Supporting TablesTable A.1: Date of isolation for CBM lines. 56 putative mutation lines were isolated from three deep-wellboxes (A-C) following exposure to copper (started on 19 Jan 2011). Culture from each well showing growthwas streaked onto a YPD plate and assessed for colony size. Eight colonies were randomly chosen from alllines that grew normally on the YPD plates and assayed for growth in copper12. A single copper-resistantcolony was chosen from each of the 34 remaining putative mutation lines. All lines were subsequentlystreaked onto YPG plates to assay respiratory capacity.Date of isolation CBM Line Box copper12 growth (# of 8) YPG growth27 Jan 2011 CBM1 A 727 Jan 2011 CBM2 B 627 Jan 2011 CBM3 B 327 Jan 2011 CBM4 B 727 Jan 2011 CBM5 C 528 Jan 2011 CBM6 A 428 Jan 2011 CBM7 A 628 Jan 2011 CBM8 A 0† petite28 Jan 2011 CBM9 B -‡28 Jan 2011 CBM10 B - petite28 Jan 2011 CBM11 B 828 Jan 2011 CBM12 B - petite28 Jan 2011 CBM13 B 828 Jan 2011 CBM14 C 728 Jan 2011 CBM15 C - petite28 Jan 2011 CBM16 C 7⇤ petite28 Jan 2011 CBM17 C 628 Jan 2011 CBM18 C 729 Jan 2011 CBM19 A - petite29 Jan 2011 CBM20 B 8⇤ petite29 Jan 2011 CBM21 B 729 Jan 2011 CBM22 C 629 Jan 2011 CBM23 C no growth on YPD -29 Jan 2011 CBM24 C 830 Jan 2011 CBM25 A 430 Jan 2011 CBM26 A 630 Jan 2011 CBM27 A -‡30 Jan 2011 CBM28 A -‡30 Jan 2011 CBM29 A 8⇤ petiteContinued on next page114A.2. Supporting TablesTable A.1 – continued from previous pageDate of isolation CBM Line Box copper12 growth (# of 8) YPG growth30 Jan 2011 CBM30 B 530 Jan 2011 CBM31 B - petite31 Jan 2011 CBM32 B - petite31 Jan 2011 CBM33 B 831 Jan 2011 CBM34 B 831 Jan 2011 CBM35 B 031 Jan 2011 CBM36 B 831 Jan 2011 CBM37 B 831 Jan 2011 CBM38 C - petite31 Jan 2011 CBM39 C 02 Feb 2011 CBM40 A - petite2 Feb 2011 CBM41 A 02 Feb 2011 CBM42 A - petite2 Feb 2011 CBM43 A - petite2 Feb 2011 CBM44 B 82 Feb 2011 CBM45 B 83 Feb 2011 CBM46 A 43 Feb 2011 CBM47 A 73 Feb 2011 CBM48 A - petite3 Feb 2011 CBM49 B 83 Feb 2011 CBM50 B 03 Feb 2011 CBM51 B 53 Feb 2011 CBM52 B 03 Feb 2011 CBM53 B 63 Feb 2011 CBM54 B 33 Feb 2011 CBM55 B 43 Feb 2011 CBM56 C no growth on YPD -⇤ These lines were included in our study because colonies were not noticeably petite on YPD.Whole-genomesequencing indicated very little depth of coverage for mitochondrial genes for the copper resistant coloniesanalysed. These were subsequently shown to be incapable of growth on YPG plates (respiration deficient).† Because colonies on YPD plates were not noticeably petite, this line was assayed for copper tolerance. Asnone of the 8 colonies grew, this line was dropped.‡ These colonies were small on YPD but later shown to be capable of respiration (growth on YPG). Whole-genome sequencing was then conducted to determine the genetic basis of copper resistance.115A.2. Supporting TablesTable A.2: T-test results comparing maximum growth rates of the CBM lines to growth of BY4741 in YPD+ ferric citrate. Maximum growth rates were highly correlated among iron concentrations (10mM vs 40mM:r = 0.89, t = 11.00, df = 31, p < 0.001; 60mM vs 40mM: r = 0.56, t = 3.72, df = 31, p = 0.0008), soonly 40mM results are presented in the text. Growth was assayed by automated OD readings over a 24 hourperiod in the Bioscreen C.10 mM ferric citrate 40 mM ferric citrate 60 mM ferric citrateLine t df p-value t df p-value t df p-valueCBM1 -0.14 6.09 0.89 -0.34 7.80 0.74 -0.83 6.35 0.44CBM2 -1.23 4.64 0.28 -5.47 6.78 0.001 1.8 4.83 0.13CBM3 -0.04 5.75 0.97 -7.01 7.81 0.0001 2.81 6.6 0.03CBM4 1.03 6.97 0.34 -2.16 5.43 0.078 0.32 6.23 0.76CBM5 -1.86 4.69 0.13 -0.03 6.16 0.98 0.89 11.04 0.39CBM6 -4.09 6.01 0.01 -8.48 5.86 0.0002 -2.59 9.58 0.03CBM7 -10.42 5.64 0 -12.66 5.49 < 0.0001 -0.67 6.12 0.53CBM11 -0.58 5.26 0.59 -9.81 4.99 0.0002 3.27 5.79 0.02CBM13 -0.19 7.99 0.86 -0.22 5.80 0.83 1.89 4.67 0.12CBM14 2.51 7.27 0.04 -0.0001 5.23 1.00 1.42 4.69 0.22CBM16 -2.76 7.88 0.03 -27.41 7.61 < 0.0001 -2.58 8.05 0.03CBM17 -8.88 4.52 0 -16.51 7.24 < 0.0001 -4.86 18.19 0CBM18 -1.41 4.85 0.22 -3.42 6.07 0.01 1.31 4.5 0.25CBM20 -6.3 4.42 0 -33.61 4.64 < 0.0001 -2.1 6.95 0.07CBM21 -9.09 4.96 0 -11.87 9.00 < 0.0001 0.29 6.68 0.78CBM22 -2.85 7.91 0.02 -11.10 5.11 0.0001 -3.74 8.64 0CBM24 1.32 4.79 0.25 -1.17 5.32 0.29 0.34 6.71 0.75CBM25 0.5 7.28 0.63 3.67 5.19 0.01 1.24 25 0.23CBM26 -15.22 7.59 0 -36.20 5.11 < 0.0001 -3.01 12.37 0.01CBM29 -4.92 7.69 0 -28.54 7.87 < 0.0001 -1.91 18.92 0.07CBM30 -0.41 5.85 0.7 -6.37 7.97 0.0002 4.36 6.03 0CBM33 -1.28 7.96 0.24 -1.49 4.83 0.20 1.59 6.38 0.16CBM34 -3.07 7.92 0.02 -2.14 5.72 0.08 1.45 6.84 0.19CBM36 -1.42 6.01 0.21 0.26 5.31 0.81 0.01 5.78 1CBM37 0.06 5.45 0.96 -0.38 5.22 0.72 -1.37 8.43 0.21CBM45 -0.26 5.69 0.8 -0.13 5.10 0.90 1.31 6.04 0.24CBM46 -0.3 7.67 0.77 -0.90 4.69 0.41 0.89 6.23 0.41CBM47 0.05 5.99 0.96 -0.82 4.95 0.45 2.45 13.5 0.03CBM49 -11.04 4.48 0 -7.83 4.72 0.0007 -0.38 5.16 0.72CBM51 0.44 7.68 0.67 -0.21 5.74 0.84 0.52 10.86 0.61CBM53 -0.01 7.97 0.99 -0.93 5.36 0.39 0.44 5.54 0.68CBM54 0.14 7.6 0.89 1.18 7.28 0.27 1.09 6.95 0.31CBM55 1.38 4.5 0.23 1.54 7.84 0.16 2.12 7.33 0.07116A.2.SupportingTablesTable A.3: Predicted transcription factor binding site gains and losses from intergenic mutations. The nearest ORFs upstream (5’ on theWatson strand)or downstream (3’) are given, as well as the distances to the start sites (brackets) and the orientation of the binding site (S: binding site precedes startof ORF; E: binding site after end of ORF). In bold are binding sites within 500 bp of the start of a gene on the coding strand. Neighbouring repeatelements, multi-copy tRNAs, or dubious ORFs were ignored.Line Position Upstream ORF Downstream ORF Mutation TF lost TF gainedCBM1 XVI.420661 YPL071C [143] (S) MUK1 [287] (S) A>T n/a n/aCBM3 VII.150650 COX4 [479] (S) TPN1 [2126] (E) G>T FKH2, FKH1, HCM1 ORC2, SFP1, SPT15CBM5 XIV.284255 YNL190W [1860] (E) SRP1 [5] (S) T>G FKH2, HCM1, SUM1 n/aNHP6A, NHP6B, ORC2PHO2, SMP1, SPT15, YAP1CMB5,13,21 X.654261 YJR124C [23] (S) ENT3 [1702] (E) T>C SUM1, ORC2, STB3 n/aCBM7 III.306327 YCR102C [860] (S) PAU3 [1474] (S) G>T n/a GAT1, GLN3CBM7 IX.370383 PAN1 [475] (S) YIR007W [321] (S) C>G n/a GAT1, GLN3, GZF3,ECM23, SRD1CBM11 XII.605283 CDC42 [496] (S) BNA5 [1836] (E) 1D indel n/a ORC2, SFP1, YGR067CCBM24 IV.805517 SEC7 [3295] (S) HSP42 [1104] (S) G>A STP4 UME6CBM24 IV.805485 SEC7 [3263] (S) HSP42 [1136] (S) A>G LYS14, YKL222C, YRR1 AR080, CEP3, PUT3,RDS2, TBS1CBM29 XV.566240 ADE2 [49] (S) AFI1 [3318] (E) G>C RAP1 n/aCBM29 VII.1376 (telomere) COS12 [1414] (S) A>C NHP6B, NHP6A, ORC2 n/aPHO2, SPT15, YOX1CBM34 XI.364516 PTM1 [1894] (S) SNR69 [260] (S) complex n/a GAT1, GLN3, GZF3,1I indel NHP6B, PHO2, SFP1CBM49 V.438349 YER134C [546] (S) GDI1 [1267] (S) G>C SKN7 n/aCBM49 XIII.420239 PDS5 [210] (S) VPS20 [1910] (E) A>C DOT6 ERT1CBM51 IV.310552 RPP1A [430] (S) THI3 [1919] (E) A>G FKH2, HCM1, NHP6A, ORC2, SUM1NHP6B, PHO2, SPT15117A.2. Supporting TablesTable A.4: T-test results comparing maximum growth rate in copper8 between CBM lines and BY4741.Growth was assayed by automated OD readings over a 24 hour period in the Bioscreen C.Maximum growth rateLine t df p-valueCBM1 13.65 25.16 < 0.0001CBM2 11.41 24.19 < 0.0001CBM3 10.60 25.60 < 0.0001CBM4 9.32 25.73 < 0.0001CBM5 14.47 24.05 < 0.0001CBM6 13.25 24.39 < 0.0001CBM7 8.80 24.11 < 0.0001CBM11 12.78 24.79 < 0.0001CBM13 14.36 24.56 < 0.0001CBM14 11.09 25.54 < 0.0001CBM16 11.44 24.71 < 0.0001CBM17 12.09 24.68 < 0.0001CBM18 15.29 25.43 < 0.0001CBM20 11.56 24.36 < 0.0001CBM21 8.12 24.19 < 0.0001CBM22 12.21 25.66 < 0.0001CBM24 0.43 5.49 0.68CBM25 9.93 25.85 < 0.001CBM26 3.81 27.14 0.0007CBM28 6.73 25.66 < 0.0001CBM29 11.02 24.65 < 0.0001CBM30 10.72 25.89 < 0.0001CBM33 14.71 24.24 < 0.0001CBM34 11.10 26.13 < 0.0001CBM36 12.53 21.12 < 0.0001CBM37 14.89 27.70 < 0.0001CBM44 14.02 26.02 < 0.0001CBM45 13.05 25.99 < 0.0001CBM46 13.78 26.47 < 0.0001CBM47 14.37 27.20 < 0.0001CBM49 7.88 26.01 < 0.0001CBM51 7.32 25.05 < 0.0001CBM53 10.24 25.84 < 0.0001CBM54 4.54 24.60 0.0001CBM55 11.30 26.39 < 0.0001118A.2. Supporting TablesTable A.5: T-test results comparing maximum growth rate in YPD between CBM lines and BY4741. Growthwas assayed by automated OD readings over a 24 hour period in the Bioscreen C.Maximum growth rateLine t df p-valueCBM1 -2.62 27.98 0.014CBM2 -1.50 6.89 0.18CBM3 -4.72 26.93 <0.0001CBM4 0.90 5.96 0.40CBM5 -1.44 18.46 0.17CBM6 -5.86 26.06 <0.0001CBM7 -2.20 5.95 0.07CBM11 -3.86 14.88 0.002CBM13 -2.31 27.00 0.028CBM14 0.0004 7.30 1.00CBM16 -6.69 25.79 <0.0001CBM17 -8.63 26.77 <0.0001CBM18 -0.66 7.18 0.53CBM20 -0.48 5.34 0.65CBM21 -0.22 5.02 0.84CBM22 -8.88 26.52 <0.0001CBM24 1.86 8.33 0.10CBM25 1.69 6.10 0.14CBM26 -19.28 26.76 <0.0001CBM28 -7.63 24.87 <0.0001CBM29 0.24 5.65 0.82CBM30 -1.80 6.04 0.12CBM33 -1.63 9.12 0.14CBM34 -3.89 26.23 0.0006CBM36 -5.19 24.35 <0.0001CBM37 -1.54 6.94 0.17CBM44 -1.50 9.98 0.17CBM45 -4.23 26.70 0.0002CBM46 -4.47 25.03 0.0001CBM47 -1.51 9.90 0.16CBM49 -9.68 25.53 <0.0001CBM51 0.20 6.21 0.85CBM53 0.30 9.01 0.77CBM54 0.46 6.10 0.66CBM55 1.96 5.78 0.10119A.2. Supporting TablesTable A.6: T-test results comparing maximum growth rate in YPG between CBM lines and BY4741. Growthwas assayed by automated OD readings over a 24 hour period in the Bioscreen C.Maximum growth rateLine t df p-valueCBM1 0.25 13.04 0.81CBM2 -5.33 16.19 0.0001CBM3 -6.68 18.30 < 0.0001CBM4 1.01 17.52 0.33CBM5 -0.30 14.22 0.77CBM6 -56.27 17.09 < 0.0001CBM7 -5.78 16.03 < 0.0001CBM11 -8.83 15.96 < 0.0001CBM13 -1.03 11.74 0.32CBM14 0.84 13.59 0.42CBM16 -59.60 18.40 < 0.0001CBM17 -12.32 12.62 < 0.0001CBM18 -0.41 21.13 0.69CBM20 -69.59 24.25 < 0.0001CBM21 -21.28 13.39 < 0.0001CBM22 -31.41 20.28 < 0.0001CBM24 -1.58 23.14 0.13CBM25 1.94 19.23 0.07CBM26 -21.35 20.68 < 0.0001CBM28 -39.78 36.21 < 0.0001CBM29 -91.71 45.36 < 0.0001CBM30 -5.56 21.67 < 0.0001CBM33 -1.64 26.01 0.11CBM34 -0.75 15.98 0.46CBM36 -0.20 13.43 0.85CBM37 -1.33 14.82 0.20CBM44 -1.71 15.21 0.11CBM45 -2.14 12.96 0.05CBM46 -1.04 16.19 0.31CBM47 -0.35 13.78 0.73CBM49 -7.57 13.07 < 0.0001CBM51 0.27 14.87 0.79CBM53 0.18 13.44 0.86CBM54 -0.61 13.69 0.55CBM55 -0.77 18.51 0.45120A.2. Supporting TablesTable A.7: Summary of linear model analyses of the maximum growth rate of tetrads, assayed in YPD withinthe Bioscreen C. For each line, we show the analysis of a full model accounting for the genes listed below.Thus, CBM2 shows evidence for an effect of an additional copy of chrII on growth in YPD. Significantp-values are in bold.Line Term Estimate t df p-valueCBM2 CUP1 -4.66E-08 -0.48 13 0.64+chrII -7.37E-02 -3.39 13 0.0048CBM14 CUP1 8.20E-08 1.60 21 0.12MAM3 -3.61E-02 -1.75 21 0.095CBM25 CUP1 2.71E-08 0.62 12 0.54MLP1 -2.67E-02 -1.086 12 0.30ENA5 5.62E-03 0.26 12 0.80CBM34 CUP1 -1.58E-08 -0.48 21 0.63VTC4 2.17E-02 1.29 21 0.21Table A.8: Summary of linear model analyses of the maximum growth rate of tetrads assayed in copper9,after correcting for growth in YPD (maximum growth rate in copper9 minus maximum growth rate in YPD),both measured within the Bioscreen C. For each line, we show the analysis of a full model accounting for thegenes listed below. Only CBM2 showed evidence for an effect of a mutation on growth in YPD (see TableA.7). Significant p-values are in bold.Line Term Estimate t df p-valueCBM2 CUP1 2.96E-07 3.18 13 0.0072chrII 1.10E-01 5.30 13 0.00015CBM14 CUP1 4.40E-08 0.80 21 0.43MAM3 5.49E-02 2.48 21 0.022CBM25 CUP1 8.50E-08 1.53 12 0.15MLP1 2.11E-02 0.67 12 0.51ENA5 -5.93E-03 -0.21 12 0.84CBM34 CUP1 1.70E-07 4.12 21 0.00049VTC4 3.48E-02 1.65 21 0.11121A.2.SupportingTablesTable A.9: Additional mutations identified in the small-colony forming CBM lines. CUP1 coverage for each line is provided in the second columnand does not account for additional copies via chrVIII aneuploidy.CUP1 Genome Position Mutation Position Amino acidCBM line coverage (chr.bp) Gene (Watson strand) (from 5’ end) change ExchangeabilityCBM9 0.69 VII.481622 PMA1a C>T 1045 Gly>Ser 0.304CBM27 0.83 VII.482121 PMA1 3D indel (AAC/—) 544 Val> mito.83071 intergenic A>GCBM28 1.25 IV.43829&IV.43830 intergenic CA>ATchrIII aneuploidychrV aneuploidychrVIII aneuploidya As a sample from the population was sequenced, this mutation was not fixed but was called as a "heterozygote" (43.4% of reads).122A.3. Supporting FiguresA.3 Supporting Figures●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24BY47410.00.51.01.5●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24BY4739● ●●●●●●●●●●●●●● ●●● ●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM20.00.51.01.5●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM2 t1d●●●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM2 t3d● ●●●●●●●●●●●●●● ●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM140.00.51.01.5●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM14 t2d●●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM14 t9d●●●●●●●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM250.00.51.01.5●●●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM25 t1c●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM25 t2b● ●●●●●●●●●●●●●● ●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM340 2 4 6 8 100.00.51.01.5●●●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM34 t3b0 2 4 6 8 10●●●●●●●●●●●●●●●●●●exp(ddn[[i]]$enviro)ddn[[i]]$ODn24CBM34 t10a0 2 4 6 8 10Concentration of copper (mM)Optical density at 24hFigure A.1: Optical density after 24 hours of growth in the Bioscreen C for specific spores over a range ofcopper concentrations. Spores were chosen that had lower CUP1 copy number and carried either an extracopy of chrII (CBM2 lines), a SNP in MAM3 (CBM14 lines), a SNP in MLP1 (CBM25 lines) or a SNPin VTC4 (CBM34 lines). Grey and black circles represent data points collected on two separate days, withtwo replicates per day. Curves drawn in red are maximum likelihood fits using the methods described inGERSTEIN et al. (2012), with the estimated IC50 represented by a vertical black line and its corresponding95% confidence interval shown by the grey dashed lines.123A.3. Supporting Figures0 1 2 3 40.51.01.52.02.5CUP1 copies (Illumina)CUP1 expression (qPCR)CBM2 CBM4 CBM13CBM14CBM16CBM22CBM24CBM34CBM37CBM51A0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.50.51.01.52.02.53.03.5CUP1 copies (Illumina)CUP1 band brightness (Southern)25:T1A25:T1B25:T1C25:T1D25:T2A25:T2B25:T2C 25:T2D25:T6A25:T6B25:T6C25:T6D25:T7A25:T7B25:T7C25:T7D2:T3A2:T3B2:T3C2:T3DBFigure A.2: Comparison of CUP1 copy number assays. A. CUP1 expression level was determined via qPCRand compared with the in silico qPCR estimates based on the fastq Illumina files for ten CBM lines. Expres-sion levels for CUP1 (normalized to TAF10) were obtained by qPCR, with the y-axis giving expression levelsrelative to the BY4741 ancestral line. The slope is significant when forced through the (1,1) point, whichassumes that both axes are scaled to the ancestor (even though the derived BMN lines and not BY4741 wereused as the control in the in silico qPCR assays; p = 0.02, solid), but the slope is not significant otherwise (p= 0.27, dashed). B. CUP1 copy number was estimated by band brightness from Southern blots and comparedwith in silico qPCR estimates for the lines established from tetrads. Band brightness from the Southerns wasnormalized to the average of two BY4739 bands run on the same gels (recall that the CBM lines had beencrossed to BY4739 to generate the tetrads; normalizing to the two BY4741 bands yielded similar results).Only those tetrads for which whole-genome sequencing was performed are included (e.g., “25:t1a” refers to“CBM25, tetrad 1, colony a”). The slope is significant when forced through the (1,1) point (p = 0.02; solid)and marginally significant otherwise (p = 0.08, dashed).124A.3. Supporting Figures●●●●●●●● ●●●●●●●●●●●●●●IC50 (mM Cu)02468BY4741boi2∆bud3∆ccp1∆clb3∆did4∆esc1∆fes1∆flo10∆grc2∆gsc2∆hmg2∆mal12∆mam3∆mlp1∆prk1∆sbe22∆smi1∆trm7∆vam6∆vtc1∆vtc4∆Figure A.3: Copper tolerance of S. cerevisiae knockout lines for genes identified in our experiment. Solidcircles identify lines that have significantly different tolerance than BY4741, measured as IC50 (bars repre-sent 95% confidence intervals). The horizontal lines are for illustrative purposes to indicate the mean (solidline) and confidence interval (dashed lines) for BY4741.125A.3. Supporting Figures● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ●●●● ●●●●3:c(length(range) + 2)0.00.10.20.30.4●●+chrIICUP1ï + ï + ï + + ï ï + ï + + + ï ï+ïBY4739 CBM2 t1 t2 t3 t5A● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●● ●●●● ●●●●●●3:c(length(range) + 2)0.00.10.20.30.4●●mam3CUP1ï + + ï ï ï + + ï + ï + ï + ï + + ï ï + + ï + ï+ïBY4739 CBM14 t1 t2 t8 t9 t10 t11B● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●3:c(length(range) + 2)0.00.10.20.30.4●●mlp1ena5CUP1ï + + ï + + ï ï + + ï ï + ï + ïï ï + + + ï ï + ï ï + + ï ï + ++ï +ïBY4739 CBM25 t1 t2 t6 t7C● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ●●●●●●●●●● ●● ●●●●●●●0.00.10.20.30.4●●vtc4CUP1+ ï + ï ï + ï + ï ï + + ï ï + + + ï ï + ï + ï ++ïBY4739 CBM34 t2 t3 t4 t6 t10 t11DMaximum growth rate in YPD (/h)min CUP1max CUP1Figure A.4: Maximum growth rate of tetrads in YPD, as measured by the Bioscreen C. Tetrads were derivedfrom four different CBM lines: A. CBM2, B. CBM14, C. CBM25, and D. CBM34. For each line, maximumgrowth rate was assayed within the Bioscreen C on a single day (±1 SE across replicate wells). The darknessof the circle represents the relative number of copies of CUP1, as assayed from Southern blots. Presence(+) or absence () of a segregating mutation is also noted. All lines are compared to the growth rate of thetwo parents, BY4739, and the relevant CBM parent (red lines), except for the tetrads derived from CBM25for which parental growth rate was not assayed (due to its initially being considered CBM22, see Materialsand Methods).126A.3. Supporting Figures0246810Copper tolerance(IC50 in mM)BY4741BY4739CBM2CBM2 t1dCBM2 t3dCBM14CBM14 t2dCBM14 t9dCBM25CBM25 t1cCBM25 t2bCBM34CBM34 t3bCBM34 t10a** * * *+chrII mam3 mlp1 vtc4Figure A.5: Copper tolerance, as measured by IC50 after 24 hours of growth in the Bioscreen C for specificspores. Lines were chosen because they had low CUP1 copy number and carried either an extra copy ofchrII (CBM2 lines), a mutation inMAM3 (CBM14 lines), a mutation inMLP1 (CBM25 lines), or a mutationin VTC4 (CBM34 lines). All mutant lines had a significantly higher IC50 than either of the BY controls,and all spores had a significantly lower IC50 than their CBM parent. Horizontal bars indicate statisticalcomparisons, where an asterisk (*) above a bar indicates statistical significance (p < 0.05). Among thespores carrying the same allele, only the CBM25 spores differed significantly from one another in IC50, andonly marginally so if corrected for multiple comparisons. Note that CBM25 t1c also carries the mutation inENA5. Vertical bars represent 95% confidence intervals.127Appendix BAppendix for Chapter 3: WidespreadGenetic Incompatibilities BetweenFirst-Step Mutations During ParallelAdaptation of Saccharomyces cerevisiae to aCommon EnvironmentB.1 Supplementary MethodsB.1.1 Strain construction detailsAll possible haploid and diploid genotypes were created for each pair of four beneficial mutations (one ineach of ERG3, ERG5, ERG6 and ERG7, Table 3.1). Each mutation was initially isolated in the BY4741haploid background (MATa his3D1 leu2D0 met15D0 ura3D0) and given a Beneficial Mutation Nystatin(BMN) strain number (GERSTEIN et al. 2012). Each BMN strain was mated to BY4739 (MATa leu2D0lys2D0 ura3D0) (Open Biosystems) to create strains heterozygous for a single ERG mutation, and diploidswere positively selected on plates lacking both histidine and lysine. Similarly, diploid non-mutant strainswere created by mating BY4741 and BY4739. In each case, single colonies were then grown up on a secondselection plate and frozen at -80°C in 15% glycerol.MATa single mutant strains were isolated by sporulation of the heterozygous diploids. Diploid stockgrown on a YPD plate was used to inoculate 10 mL of YPD and grown overnight on a rotor at 30°C. 200µL of culture was then washed, spread on potassium acetate plates (1% KOAc, 2% agar) and sporulatedat 25°C until a sufficient number of tetrads could be observed. The resulting tetrads were dissected bymicromanipulation on YPD plates. The spores were allowed to germinate and grown at 30°C for threedays before replica plating to test for auxotrophies, mating type, and nystatin growth ability. Auxotrophywas assessed on SC plates lacking the appropriate amino acid. Mating type was tested by replica platingtetrads onto plates containing a lawn of MATa or MATa yeast carrying a histidine (his1-123) auxotrophy,allowing them to mate, and subsequently testing for mating on a plate lacking arginine, histidine, leucine,lysine, methionine, tryptophan, adenine and uracil (i.e., a plate on which no original haploid strain couldgrow). Nystatin growth was assessed on YPD + 8 mM nystatin because growth of ancestral strains wasnot noticeably inhibited on plates with a lower concentration of nystatin. YPD + 8 mM nystatin plates128B.1. Supplementary Methodswere made by preparing YPD medium with agar as usual, subsequently adding the appropriate amountof 2.7 mM nystatin stock, and mixing by inversion immediately before pouring. All tetrads were verifiedfor 2:2 segregation of auxotrophies and mating type. Once this was confirmed, the spores that showedgrowth on the nystatin plate and contained the desired MATa lys2D0 mutation were frozen at -80°C in 15%glycerol. Throughout strain construction, histidine and lysine auxotrophies were consistently kept with thesame mating types so that all haploid strains were either MATa his3D1 or MATa lys2D0. The methionineauxotrophy (met15D0) did not show strong selection on plates lacking methionine and was not tracked.The haploidMATa strains were then mated to the originalMATa strains to create strains that were eitherhomozygous for one mutation or heterozygous for two mutations. Diploids were selected and frozen asdescribed for the singly heterozygous strains.The haploid double mutant strains were created through sporulation and dissection of the double het-erozygous strains. Three strains (erg3/ERG3 erg6/ERG6, erg3/ERG3 erg7/ERG7, and erg6/ERG6 erg7/ERG7)were struck from frozen on YPD plates and grown at 30°C for 2-3 days. They were sporulated, dissected andchecked as described above except that they were moved to 20°C after three days of sporulation. This proto-col was repeated for erg3/ERG3 erg5/ERG5, except that it was kept at 20°C from the beginning. erg5/ERG5erg7/ERG7 would not sporulate under these conditions. To obtain the MATa double mutant strain it wassporulated in 10 mL 1% KOAc + amino acids liquid medium at 20°C. In order to obtain the MATa doublemutant strain, it was sporulated by streaking a patch of cells onto a GNA pre-sporulation plate (5% dextrose,3% nutrient broth, 1% yeast extract, 2% agar) and growing at 30°C overnight, repeating the streaking andgrowth on another GNA pre-sporulation plate, and finally sporulating in 2 mL of supplemented sporulationmedium (1% potassium acetate, 0.005% zinc sulphate, 2 mg/100 ml uracil, 10 mg/100 ml leucine) on arotor at 25°C for five days, then moving to 30°C until a sufficient number of tetrads were found. The sameprocedure was applied to erg5/ERG5 erg6/ERG6 to obtain theMATa double mutant strain. Tetrads were cho-sen that showed 2:2 segregation of the nystatin resistance (assessed either on YPD + 8 mM nystatin plates,YPD + 10 mM nystatin plates or in a liquid assay), indicative of two double mutant spores and two wildtypespores. Double mutant strains were frozen at -80°C in 15% glycerol. All haploid double mutant strains wereconfirmed by Sanger sequencing.We failed to obtain the MATa erg5 erg6 double mutant strain through crossing and sporulation becausethe two genes are linked (they are 48 kb apart but flank the centromere of chr XIII). For this strain, trans-formations were performed using a protocol based on CREGG (2007). MATa erg5 yeast were grown from asingle colony in 10 mL YPD at 30°C. The next day, two new 10 mL YPD tubes were inoculated with 500 mLof yeast from the overnight culture and grown at 30°C until reaching an OD600 between 0.5 and 0.6. One tubewas used for the transformation, and one was used as a negative control. Cells were collected by spinningthe cultures down for 5 minutes at 4500 rpm and were washed twice with water using a spin of 10 minutesat 4500 rpm. The yeast were resuspended in 2 mL of cold 1 M sorbitol, spun at 5000 rpm long enough topellet the cells, the supernatant was removed, and the yeast were resuspended in 1 mL of cold 1 M sorbitol.80 mL of these cultures were then electroporated, along with either 8 mg of an oligonucleotide designed tocontain the ERG6 SNP of interest (sequence: TTCAAAGAGGCGATTTAGTTCTCGACGTTCGTTGTG-GTGTTGGGGGCCCAGCAAG) or an equal volume of water, using a BioRad Gene Pulser Xcell and the129B.1. Supplementary Methodsparameters defined in CREGG (2007). Immediately after electroporation, 1 mL of YPD was added to theyeast and the cells were incubated for 1 hour at 30°C to recover. The cells were then plated on YPD + 10mM nystatin plates and incubated at 30°C until colonies were visible. The insertion of the mutation in erg6was verified by Sanger sequencing.Strains with one heterozygous and one homozygous locus as well as double homozygous strains werecreated by mating the corresponding single mutant strains orMATa double mutant strains to theMATa doublemutant strains, as described above.B.1.2 Segregating mutation in DSC2The original strain with a mutation in ERG7 also carried a second mutation in the gene DSC2 (Table 3.1).This mutation was not originally tracked when constructing the strains and it was later identified by Sangersequencing in all haploid strains constructed from the original strain carrying a mutation in ERG7 (Ta-ble B.1). Two combinations of strains that differed in their status at DSC2 between the mating types (erg7and erg5 erg7) and were tested for differences in maximum growth. No significant difference was found forgrowth rate in nystatin whether we treat each replicate as independent or average data points collected on thesame day (Welch two sample t-tests with replicates treated as independent, erg7: t = -0.38, df = 49.56, P =0.71; erg5 erg7: t = -1.01, df = 37.66, P = 0.32). In YPD, a significant difference was found only for the genecombination erg5 erg7 and only when all replicates were treated as independent (Welch two sample t-tests,erg7: t = 0.27, df = 50, P = 0.79; erg5 erg7: t = -2.32, df = 35.97, P = 0.026). The test was not significantwhen data points for each day were averaged (erg5 erg7: t = -1.90, df = 7.87, P = 0.094). Furthermore, thedifference between mutant and wildtype DSC2 growth rates was in each case minor and did not substantiallyalter the data points illustrated in main text Fig 3.3 or the conclusions drawn.Table B.1: DSC2 allele status in haploid strains constructed from the original strain carrying a mutation inERG7.Strain Mating Type Allele status at DSC2erg7 a mutanterg7 alpha wildtypeerg3 erg7 a mutanterg3 erg7 alpha mutanterg5 erg7 a mutanterg5 erg7 alpha wildtypeerg6 erg7 a wildtypeerg6 erg7 alpha wildtypeB.1.3 Preparing stocks for growth rate assaysA total of seven growth rate assays were conducted for our analysis of epistasis. We had originally intendedto perform three assays, but four more were performed to maintain the intended level of replication after130B.1. Supplementary Methodsencountering problems with growth and strain construction. For an overview of which lines were included inwhich fitness assays, see S1 Table and for complete information about the growth assays, see files depositedat Dryad. Many of the lines involved in this study had poor growth even in a rich medium. Because of this,care was taken to standardize initial cell densities (“pre-assays”) for use in subsequent growth rate assays.General methods will be explained first, with exceptions to these methods explained subsequently.The pre-assay took place in 100-well honeycomb Bioscreen plates using a permissive medium of 148.5mL of YPD + 0.5 mM nystatin (except for the first assay, which used only YPD). YPD + 0.5 mM nystatin wasused to help prevent reversion of strains with severe growth defects in YPD while still permitting the growthof all strains. The wells were inoculated with 1.5 mL of frozen culture. Replicates were randomized withinplates, always including all lines on the same plate for a given pair of mutations. The plates were incubatedin the Bioscreen C Microbiological Workstation at 30°C with maximum continuous shaking, measuring theoptical density (OD) of the cultures every 30 minutes using the wideband filter.The cultures were incubated inthis way for 72 hours, which was enough time for most strains to obtain clear growth (defined as a maximumOD of about two times the initial OD); anything below this threshold was excluded from analysis unlessotherwise noted. Maximum OD was used to determine the volume to transfer for the growth rate assays. If itwas above 1, we transferred 1.5 mL into one plate containing 148.5 mL of YPD and one plate containing 148.5mL of nystatin2 (using the same randomized well map). If the maximum OD was below 1, it was rounded tothe nearest 0.05, and the transferred volume was scaled accordingly (giving final volumes ranging between150 mL and 156 mL).To investigate whether the pre-growth medium influenced growth rate, we ran a sign test comparing themean maximum growth rates in the nystatin2 assay between Assay 1 (in which all strains were pre-grownin YPD) and Assay 2 (in which all strains were pre-grown in 0.5 mM nystatin). All 47 strains that wereincluded in both assays (and not omitted due to growth problems) were included in the sign test, which wasrun using the function binom.test in the package stats by counting the number of strains for which maximumgrowth rate was higher in Assay 2 and comparing that to what is expected by chance (p = 0.5). No significantdifference was found (P = 0.56; similar results were obtained with a paired t-test: P = 0.24).The erg6/erg6 erg7/erg7 diploid strain showed consistently poor growth, and all of the data for this straincomes from the fourth and sixth assays where the pre-assay was conducted over a longer period of time in alarger volume of liquid in an attempt to initiate the assays with the same number of cells. Briefly, 10 mL of0.5 mM nystatin in a test tube was inoculated with 15 mL of erg6/erg6 erg7/erg7 from frozen two days beforeall other lines were inoculated from frozen. The tube was incubated at 30°C on a rotor for this time. On theday when all other strains were being inoculated from frozen, the 10 mL tube of erg6/erg6 erg7/erg7 wasspun down in multiple 1.5 mL tubes and concentrated into 500 mL in one tube. 150 mL of this concentratedculture was used to fill the appropriate wells of the pre-assay plate. Despite this extra growth time andconcentrating of cells, erg6/erg6 erg7/erg7 still did not grow to an OD above the threshold at the end of thepre-assay in one of the two cases where growth rate was measured for this line and only barely did so in theother. Yeast was added to the assay plates from these wells according to their measured OD after the growthphase even though the OD was below the threshold (up to 7.5 mL was transferred).We also modified growth conditions for three other strains that showed poor growth in early pre-assays131B.1. Supplementary Methods(MATa erg6 erg7, MAT↵ erg6 erg7, and erg6/erg6). Once low growth from frozen was established, 2 mL(rather than 1.5 mL) of frozen stock was used to inoculate the wells in the pre-assay plates. Backup tubeswere also grown for these strains and described when used. In all cases, backup tubes that contained YPD +0.5 mM nystatin as the growth medium were inoculated from frozen at the same time as the pre-assay platesand were incubated at 30 °C, shaking at 200 rpm.In the pre-assay for the second growth rate assay, two out of four replicates of MAT↵ erg6 erg7 had stillnot grown to an OD above the threshold by 72 hours. One well was omitted. For the other well, 1 mL froma 10 mL backup tube (originally inoculated with 10 mL of frozen culture) was spun down at 3000 rpm for 3minutes, and this concentrated culture was used to replace the 150 mL on the growth plate. New OD readingswere taken, and the new OD was within the range measured for the other strains.In the third pre-assay, four out of four MAT↵ erg6 erg7 wells were below the threshold for detectinggrowth after 72 hours. The liquid from the wells was replaced with culture from four 10 mL backup tubes(originally inoculated with 10 mL of frozen culture). After measuring the OD of these wells, one well wasstill not above the threshold; to ensure that enough cells were transferred for that one line, we concentratedthe cells found in 1 mL of the culture from the corresponding tube by spinning them down using a tabletopcentrifuge and removing most of the supernatant, leaving ~200 mL of concentrated culture. 1.5 mL of thisculture was transferred directly to the honeycomb plate for the growth assay.In the sixth pre-assay, one replicate of MAT↵ erg6 erg7 remained below the threshold for growth after72 hours. The liquid from the well was replaced with culture from a 1.5 mL backup tube containing 500 mLof culture (originally inoculated with 5 mL of frozen culture). A new OD reading was taken of that well andwas within the range measured for the other strains.Following each pre-assay, growth rate assays were conducted in both YPD and YPD + 2 mM nystatin(‘nystatin2’), as described in the main text.B.1.4 Analysis including outliersAll qualitative relationships between strains and the main conclusions were insensitive to the exclusion orinclusion of the identified outliers, with two exceptions for the haploids in nystatin2 (see Fig B.6 and FigB.7 for versions of Fig 3.3 and Fig 3.4 that include all outliers). One exception is that the erg3 erg5 strainno longer had a significantly lower maximum growth rate than the erg3 strain in nystatin2. This was due toone large outlier in theMATa erg3 data, which exhibited almost no growth (maximum growth rate of 0.038),while all remaining wells (including both mating types) showed substantial growth (maximal growth rateranged from 0.16 to 0.25 across 35 wells). The exclusion of this single outlier leads to the observation of asignificant difference between the aforementioned strains.A similar failure of one well to show substantial growth was observed in erg6/erg6 and erg3/erg3erg6/ERG6. In addition, two wells of erg6/ERG6 showed substantially higher growth (0.17 and 0.19), com-pared to all remaining wells (0.0022 to 0.072 across 22 wells), although our outlier exclusion algorithm onlyallowed one point to be excluded per strain. These other examples did not affect the statistical results butsuggest either occasional contamination or mutation.The other statistical difference is that the erg3 erg6 strain no longer had a significantly lower maximum132B.2. Supporting Tablegrowth rate than the erg6 strain in nystatin2. The difference between these strains is only slightly significantin the model excluding all outliers (P = 0.047) and becomes marginal when either including all outliers (P =0.083) or excluding only the one outlier replicate of erg3 (P = 0.058). We believe that this represents a lackof power to detect a true, small difference in the haploids as this relationship is supported in the homozygousdiploids (excluding outliers: P = 0.0069; including outliers: P = 0.041). For the full alternative analysiswithout outlier removal, see ONO et al. (2016).B.2 Supporting TableTable B.2: Experimental design of the growth rate assays. In each epistasis assay, growth rate wasmeasured in a Bioscreen C over a 24 hour period for a pair of ergosterol mutations (first column) usingtwo replicate wells for each genotype (ancestral, single mutant, double mutant for haploids, includingthose heterozygous or homozygous for diploids), with the exception of double mutant haploids, whichwere measured in four replicate wells. Checkmarks indicate that all data from this assay was used whilebullets indicate that some strains were omitted (see footnotes).MutationPair Assay 1 Assay 2a Assay 3 Assay 4 Assay 5 Assay 6 Assay 7erg3 & erg5erg3 & erg6erg3 & erg7erg5 & erg6 MATa onlyb MATa onlyb MATa onlyb •c •c •c derg5 & erg7 •e •e •e ferg6 & erg7 •g •g •g ha The Bioscreen bulb burned out during the pre-assay; OD readings taken after replacing the bulb were used to estimatethe final cell densities for inoculation of the growth rate assays.b Others not assayed becauseMATa erg5 erg6 was unavailable.c erg5/erg5 erg6/erg6 data from this assay were omitted from final analyses (see d).d Because the double homozygous mutant displayed high levels of growth in both YPD and nystatin, we were concernedthat the stock might contain a mixture of resistant and non-resistant cells. We thus struck this stock down to colonieson a YPD plate, picked five colonies, and used the resulting five stocks to assay growth. Each stock was confirmedto be erg5/erg5 erg6/erg6 by Sanger sequencing. For each of these stocks, two replicate wells (1 stock) or three wells(4 stocks) were used to measure growth. All mutant and non-mutant combinations of erg5 and erg6 (haploid anddiploid) were also regrown in two replicate wells in this assay. The growth rates of the diploid double mutants did notdiffer qualitatively from previous assays in nystatin but were substantially lower in YPD, consistent with the populationanalyzed in previous assays having been polymorphic (allowing non-resistant cells to proliferate). Thus, only data fromAssay 7 was used for the growth rate of erg5/erg5 erg6/erg6, although the qualitative results in nystatin are unaffectedif all data were used. Data from each well were treated independently in the analysis, given that each was grownseparately from frozen.e erg5/erg5 erg7/ERG7 data from this assay were omitted from final analyses due to an error in the creation of the originalline (identified by Sanger sequencing).f Data not collected due to Bioscreen machine error resulting in lower replication for the line erg5/erg5 erg7/ERG7.g erg6/erg6 erg7/erg7 and some erg6 erg7 haploid data could not be used due to insufficient starting cell densities of thedouble mutant lines from the pre-assays so that appreciable growth was never observed.h Data not collected due to machine error. Epistasis was large and easy to detect for this gene pair, despite the lowerreplication.133B.3. Supporting FiguresB.3 Supporting FiguresFigure B.1: Optical density after 24 hours of growth for haploid strains in nystatin2 (above diagonal) andYPD (below diagonal), plotted on a log scale. Points are the fitted least-squares means of the ODs, deter-mined in the mixed-effects model run using log(OD). ⇥’s denote the additive fitness null expectation for thedouble mutant, i.e., with no epistasis. Each single mutant is coloured differently, the double mutant is shownin black, and the ancestor is grey. Vertical bars represent 95% confidence intervals of the fitted least-squaresmean. Solid lines indicate significant comparisons, while dotted lines are non-significant comparisons. Com-binations showing significant sign (S) and reciprocal sign (RS) epistasis are indicated by the presence of theabbreviation at the top of the panel. The same outliers were removed as in the analysis of maximum growthrate because their growth rates indicate a potential problem with the replicate. Sign epistasis is less oftendetected in this analysis of log(OD) in nystatin, likely because even slower growing strains are given timeto catch up in cell density over 24 hours. All underlying raw data and analyses can be found in ONO et al.(2016).134B.3. Supporting FiguresFigure B.2: Optical density after 24 hours of growth for diploid strains in nystatin2 (above diagonal) andYPD (below diagonal), plotted on a log scale. Points are the fitted least-squares means of the ODs, withclosed circles determined in the mixed-effects model run using log(OD) including only homozygous strainsand open symbols from the model that includes heterozygous strains (open diamonds: double heterozygotes;open triangles: single heterozygotes that are wildtype at the other gene; open circles: single heterozygotesthat are homozygous mutants at the other gene). Points and bars are otherwise as in Fig B.1. All symbolsare coloured intermediately according to genotype and arrayed along the x-axis so as to lie between thetwo strains that are genotypically most similar to it. Solid lines indicate significant comparisons in testsrun including only homozygous strains while dotted lines are non-significant comparisons. See Fig B.1 forfurther graphical details. The same outliers were removed as in the analysis of maximum growth rate becausetheir growth rates indicate a potential problem with the replicate. Sign epistasis is less often detected in thisanalysis of log(OD) in nystatin, likely because even slower growing strains are given time to catch up in celldensity over 24 hours. Note that the strain erg5/ERG5 erg6/erg6 was later found to be homozygous for themutation in ERG5, likely due to a loss of heterozygosity event. All underlying raw data and analyses can befound in ONO et al. (2016).135B.3. Supporting FiguresFigure B.3: Optical density after 24 hours of growth for homozygous diploids in a range of concentrationsof nystatin. These results are qualitatively similar to the haploid strains with the exception of the erg6/erg6erg7/erg7 double mutant, which has very low growth in all concentrations of nystatin. Colours go from red topurple, through blues, from lowest to highest concentrations of nystatin. Lines connect different mutants inthe same concentration of nystatin. Differences in OD between mutants were not tested statistically and areall represented by solid lines (in contrast to Fig 3.5). Arrows on the y-axes indicate the OD of the ancestralstrain. All replicates were averaged, and error bars denote the standard error. Note that tolerance was assayedin the erg5/erg5 erg6/erg6 homozygous double mutant before we determined that it was likely polymorphic;these points may thus be underestimates (see Table B.2 for details). All underlying raw data and analysescan be found in ONO et al. (2016).136B.3. Supporting FiguresFigure B.4: Maximum growth rate of diploid strains for each gene combination in nystatin2. Genotypeat each of the two genes combined is represented along the x- and y-axes, with the ancestral genotype inthe lower left corner and the homozygous double mutant genotype in the upper right corner. Least-squaresmeans of maximum growth rates, as determined from a model including all possible diploid genotypes, arerepresented by the darkness of the boxes. Arrows indicate significant differences between genotypes, witharrowheads pointing to the significantly higher growth rate as determined by pairwise comparisons correctedfor multiple comparisons using the multivariate t distribution in lsmeans, as was done for the haploids andhomozygous diploids. Only adjacent genotypes on the grid (horizontal and vertical) were compared, withthe exception of the double heterozygous strain (centre), which was compared to all other genotypes. Notethat the strain erg5/ERG5 erg6/erg6 was later found to be homozygous for the mutation in ERG5, likely dueto a loss of heterozygosity event. All underlying raw data and analyses can be found in ONO et al. (2016).137B.3. Supporting FiguresFigure B.5: Optical density after 24 hours of growth for diploid strains in a range of concentrations of nys-tatin. Colours go from red to purple, through blues, from lowest to highest concentrations of nystatin. Linesconnect different mutants in the same concentration of nystatin. Mutant strains are ordered one mutationalstep apart along the x-axis, with the homozygous double mutant at both ends. Sections shaded in grey rep-resent mutants carrying at least one homozygous mutation. Differences in OD between mutants were nottested statistically and are all represented by solid lines (in contrast to Fig 3.5). Arrows on the y-axes indicatethe OD of the ancestral strain. All replicates were averaged, and error bars denote the standard error. Notethat tolerance was assayed in the erg5/erg5 erg6/erg6 homozygous double mutant before we determined thatit was likely polymorphic; these points may thus be underestimates (see Table B.2 for details). Also notethat the strain erg5/ERG5 erg6/erg6 was later found to be homozygous for the mutation in ERG5, likely dueto a loss of heterozygosity event. All underlying raw data and analyses can be found in ONO et al. (2016).138B.3. Supporting FiguresFigure B.6: Maximum growth rate of haploid strains in nystatin2 (above diagonal) and YPD (below diagonal)when including outliers. Points are the fitted least-squares means of the maximum growth rates, determinedin the mixed-effects model. ⇥’s denote the additive fitness null expectation for the double mutant, i.e., withno epistasis. Each single mutant is coloured differently, the double mutant is black, and the ancestor is grey.Vertical bars represent 95% confidence intervals of the fitted least-squares mean. Solid lines indicate sig-nificant comparisons, while dotted lines are non-significant comparisons. Combinations showing significantsign (S) and reciprocal sign (RS) epistasis are indicated by the presence of the abbreviation at the top of thepanel. All underlying raw data and analyses can be found in ONO et al. (2016).139B.3. Supporting FiguresFigure B.7: Maximum growth rate of diploid strains in nystatin2 (above diagonal) and YPD (below diagonal)when including outliers. Points are the fitted least-squares means of the maximum growth rates, with closedcircles determined in the mixed-effects model including only homozygous strains and open symbols fromthe model that includes heterozygous strains (open diamonds: double heterozygotes; open triangles: singleheterozygotes that are wildtype at the other gene; open circles: single heterozygotes that are homozygousmutants at the other gene). Points and bars are otherwise as in Fig 3.3 and Fig B.6. All symbols are colouredintermediately according to genotype and arrayed along the x-axis so as to lie between the two strains thatare genotypically most similar to it. Solid lines indicate significant comparisons in tests run including onlyhomozygous strains while dotted lines are non-significant comparisons. See Fig 3.3 or Fig B.6 for furthergraphical details. Note that the strain erg5/ERG5 erg6/erg6 was later found to be homozygous for themutation in ERG5, likely due to a loss of heterozygosity event. All underlying raw data and analyses can befound in ONO et al. (2016).140Appendix CAppendix for Chapter 4: The limit toevolutionary rescue depends on ploidy inyeast exposed to nystatinC.1 Strain DifferencesDespite not being judged as putative mutants, many diploid wells did show growth in the initial acquisitionexperiments. There was a significant association between the identity of a strain and whether or not it grewin the acquisition experiments (2 contingency test using chisq.test in the R package stats [R CORE TEAM2016]: Acquisition Experiment 1: 2 = 61.58, df = 2, p-value = 4.26⇥ 1014; Acquisition Experiment 2:2 = 72.7, df = 2, p-value < 1015; Acquisition Experiment 3: 2 = 168.38, df = 2, p-value < 1015). Inthe first two acquisition experiments (performed in the BY strains and YPDnystatin4),MAT↵ grew the most,proportionally, followed by MATa and the diploids, which had similar growth. In Acquisition Experiment 3(performed in the W303 strains and SCnystatin4), MATa grew the most followed by the diploids and bothgrew much more than theMAT↵ wells. This difference in growth was observed despite the additional oppor-tunity given to 80 of the MAT↵ wells that were sampled and run in duplicate in the acquisition experiment.The difference in growth of the MAT↵ wells was the main observed inconsistency between the two geneticbackgrounds used (BY vs. W303), and can be accounted for by the respiratory-deficiency of our copy of theMJM36 strain. The associations remained significant when we grouped all strains of a single ploidy, withhaploids growing more often than diploids in the first two acquisition experiments but not the third (becauseof the poor growth of MAT↵).However, there was a difference in the distribution of days until growth between strains (2 contingencytest using ‘chisq.test’ with a simulated p-value based on 10,000 replicates: Acquisition Experiment 1: 2 =49.82, p-value < 1.00⇥ 104; Acquisition Experiment 2: 2 = 90.69, p-value < 1.00⇥ 104; AcquisitionExperiment 3: 2 = 159.86, p-value < 1.00⇥ 104) (Fig. C.2), and these associations remained significantwhen we grouped all strains of a single ploidy. MATa populations had the lowest mean number of days untilgrowth, followed by MAT↵ and then the diploids (details of comparisons in Table C.1). In the follow-upgrowth assays, OD72 (and therefore whether a population was judged to be putatively nystatin-resistant) wascorrelated with number of days until growth, decreasing with increasingly later day of acquisition (Kendall’srank correlation using cor.test in the R package stats [R CORE TEAM 2016], Experiment 1: ⌧ = -0.29, z =-5.90, p-value = 3.62⇥ 109; Experiment 2: ⌧ = -0.45, z = -7.80, p-value = 6.04⇥ 1015; Experiment 3:⌧ = -0.46, z = -9.52, p-value < 1015, Fig C.2). We conclude that later-growing strains are less likely to be141C.2. Mutant Coveragetrue mutants. These wells may instead be growing due to the degradation or inactivation of nystatin in themedium over time (see Section 4.3.2).C.2 Mutant CoverageThe following Mathematica package was used to carry out the calculations and is available upon request.142Diploidy limits evolutionary rescue in yeast exposed to nystatinSupplementary Mathematica file for Ono et al.Chance that a mutation was sampled across the acquisition experimentsü Deep Well BoxesHere we model the growth of a population from a single cell established on an agar plate (rich medium), picked as acolony, and grown to saturation in 150uL rich medium, where cell densities were estimated by hemacytometer to be~7.03 µ 107cells/mL for the BY “source” population (~7.42 µ 107cells/mL for the W303 “source” population), fromwhich a sample of 10uL is taken to establish the population in each deep well.We then calculate the expected number of mutations that would occur within at least one diploid cell within each deepwell and across the entire experiment. To be conservative, we use the lower reported mutation rate per basebair of 1.67× 10-10 from Zhu et al. (2014, PNAS), rather than the higher 3.3 × 10-10 from Lynch et al. (2008, PNAS).Parameters:m = mutation rate per basepair per cell divisionN1 = population size at saturation in 150uL rich medium for growth in bioscreenf = fraction of population sampled to found a lineage (0.0667 = 10 ul/150 uL)L = # diploid deep wells (619: {191-13,286,155} in the three acquisition experiments in deep well boxes)ORF = average length of an ORF (1385 from Hurowitz, E. H., & Brown, P. O. 2003 Genome Biology)trymZHU = 1.67 µ 10-10;trymLYNCH = 3.3 µ 10-10;tryNBY = 7.03 * 107 * 0.15;H*Estimated population density per mL for BY and scaling to 150uL YPD.*LtryN303 = 7.42 * 107 * 0.15;H*Estimated population density per mL for W303 and scaling to 150uL SC.*Ltryf = 0.0667;tryL = 8191 - 13, 286, 155<; H*Number of diploid wells per acquisition experiment,excluding ones later found to have been haploid contaminated*LtryORF = 1385;Number of cell cycles required to produce source population of N1 cells:cycles = Log@2, N1DLog@N1DLog@2DIn cases where we are performing numerical sums and require an integer number, we round down the number of cycles(rounding down is slightly conservative):tryc = Floor@cycles ê. N1 Ø tryNBYD23This is the same integer number of cycles for the W303 strain:Floor@cycles ê. N1 Ø tryN303D23Total number of cell divisions involved (1 cell division from 1 Ø 2 cells, 2 cell divisions from 2 Ø 4 cells, etc):divisions = Sum@2^i, 8i, 0, cycles - 1<D-1 + N1For example, to go from 1 Ø 4 cells involves a total of 3 divisions (Ø 8 cells would involve 7 dividing cells: one 1Ø2,two 2Ø4, and four 4Ø8):divisions ê. N1 Ø 84, 8<83, 7<For the first two acquisition experiments:divisions ê. N1 Ø tryNBY êê N1.0545 µ 107The chance that NONE of these cell divisions produced a mutation at a particular site in a diploid (bearing 2m muta-tions per cell division across the two homologues) is:H1 - 2 mLdivisions;Given that m is small, this is very nearly:nomutantsinsource@m_, N1_D = ‰-2 m*N1;Using a per-basepair mutation rate of 1.67 × 10-10 (Zhu et al. 2014), the chance of no mutations at a single site withinthe source pool for a single well would be:nomutantsinsource@trymZHU, tryNBYD0.996484The distribution of mutant cell numbers in the source population is broad and very skewed (a “jackpot distribution”),and it is possible that the mutation hit early and generated many mutant cells.  To account for this mutational distribu-tion, the probability that a mutation at one specific site occurs in the jth cell cycle (going from m = 2j-1  cells to 2jcells)  is:prob@j_, m_D = 1 - H1 - 2 mLm ê. m Ø 2j-1;based on one minus the probability that no mutation hits.  (Technically, this allows for the possibility that more thanone hit would occur at the exact same site in different cells in the same cell cycle, but the chance is exceedinglyunlikely.)If a mutation does occur in the jth cycle (i.e., among the 2j cells that result in this cycle, where one is a new mutant),the fraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:frac@j_D = 1 ë 2j;Thus, prob[j,m] gives us the probability distribution for the fraction, frac[j], of mutant cells in the source population(amounting to a number of mutant cells N1 ë 2j), with the probability of at least one mutant cell at a particular siteequalling:mutantprob@m_, N1_D = Sum@prob@j, mD, 8j, 1, cycles<D‚j=1Log@N1DLog@2D I1 - H1 - 2 mL2-1+jMFor the mutation rate of Zhu et al. and the population size estimated for BY, the probability of at least one mutant cellin the source population is:2   MutantCoverage.nbmutantprob@trymZHU, tryNBYD0.00280049The expected number of mutant cells is the chance of a mutation occurring at cycle j and the number of cells that result(N1*frac[j]), summed over all cell cycles:Sum@prob@j, trymZHUD * tryNBY * frac@jD, 8j, 1, tryc<D0.0405009Finally, we calculate the probability that one or more mutant cells will be placed in a well of a deep well box, giventhat a fraction, f, of the N1 cells were sampled:probhit@m_, N1_, f_, c_D =1 - H1 - mutantprob@m, N1DL - SumAprob@j, mD * H1 - frac@jDLf*N1 , 8j, 1, c<EH*We calculate the probability of sampling some mutantcells as one minus the probability of sampling none,either because no mutations occur H1-mutantprob@m,N1DL or becausemutations occur in cycle c but are not sampled Hthe sumL*L‚j=1Log@N1DLog@2D I1 - H1 - 2 mL2-1+jM - ‚j=1c I1 - 2-jMf N1 I1 - H1 - 2 mL2-1+jMprobhit@trymZHU, tryNBY, tryf, trycD0.000555156Given L independent deep wells (each started from a different colony), where each lineage is started with a fraction, f,of its own source population,  the probability that at least one mutant cell will be sampled into at least one of the deepwell populations would be:1 - H1 - probhit@m, N1, f, cDLL1 - 1 - ‚j=1Log@N1DLog@2D I1 - H1 - 2 mL2-1+jM + ‚j=1c I1 - 2-jMf N1 I1 - H1 - 2 mL2-1+jM L1 - H1 - probhit@trymZHU, tryNBY, tryf, trycDLTotal@tryLD0.290885The expected number of deep wells with at least one mutant cell bearing a mutation at the focal site is then:Total@tryLD * probhit@trymZHU, tryNBY, tryf, trycDH*Expected number of wells with mutations.*L0.343642This is only slightly larger if we account for the higher estimated population size for W303:Sum@tryL@@iDD * probhit@trymZHU, tryNBY, tryf, trycD, 8i, 1, 2<D +tryL@@3DD * probhit@trymZHU, tryN303, tryf, trycD0.346963But the estimate would almost double if the mutation rate were closer to that inferred by Lynch et al. (2008):Total@tryLD * probhit@trymLYNCH, tryNBY, tryf, trycDH*Expected number of wells with mutations.*L0.678879Returning to our defaults, if we multiply by the average ORF length, we would expect 479 mutations per ORF to havearisen across all of the deep wells:MutantCoverage.nb   3Total@tryLD * probhit@trymZHU, tryNBY, tryf, trycD * tryORF475.944Of these mutations per ORF, we expect each gene would have, on average, 23 premature stop codons somewhere inone  of  the  wells  (fraction  of  non-sense  mutations  in  a  gene  estimated  in  Gerstein  et  al.  (2015)’s  SupplementaryMathematica package, available on DRYAD, DOI: https://doi.org/10.5061/dryad.5gp25):Total@tryLD * probhit@trymZHU, tryNBY, tryf, trycD * tryORF * 0.048831111111111124`23.2409Similarly,  Each gene would have,  on average,  350 AA changes somewhere in one of  the wells  (fraction of  non-synonymous mutations in a gene estimated in Gerstein et al. (2015)’s Supplementary Mathematica package, availableon DRYAD, DOI: https://doi.org/10.5061/dryad.5gp25):Total@tryLD * probhit@trymZHU, tryNBY, tryf, trycD * tryORF * 0.7363755555555553`350.473ü FlasksHere we model the growth of a population from a single cell established on an agar plate (YPD), picked as a colony,and grown to saturation in 10mL YPD, which corresponds to a population size of ~7µ108cells (“source” population),from which a sample of 1mL is taken to establish each individual lineage (the “founding” population).Other parameters as above.trymZHU = 1.67 µ 10-10;tryNBY = 7.03 * 107 * 10;H*Using estimated population density per mL and scaling up to 10mL*Ltryf = 0.1; H*1mL into 10mL*LtryL = 10;H*Number of diploid flasks*LtryORF = 1385;Number of cell cycles required to produce source population of N1 cells:cycles = Log@2, N1DLog@N1DLog@2DIn cases where we are performing numerical sums and require an integer number, we round down the number of cycles(rounding down is slightly conservative):tryc = Floor@cycles ê. N1 Ø tryNBYD29Total number of cell divisions involved (1 cell division from 1 -> 2 cells, 2 cell divisions from 2 -> 4 cells, etc):divisions = Sum@2^i, 8i, 0, cycles - 1<D-1 + N1The chance that NONE of these cell divisions produced a mutation at a particular site in a diploid (bearing 2m muta-tions per cell division across the two homologues) is:H1 - 2 mLdivisions;The distribution of mutant cell numbers in the source population is broad and very skewed (a “jackpot distribution”),and it is possible that the mutation hit early and generated a lot of mutant cells.  To account for this mutational distribu-tion, the probability that a mutation at one specific site occurs in the jth cell cycle (going from m = 2j-1  cells to 2jcells)  is:prob@j_, m_D = 1 - H1 - 2 mLm ê. m Ø 2j-1;based on one minus the probability that no mutation hits.  (Technically, this allows for the possibility that more thanone hit would occur at the exact same site in different cells in the same cell cycle, but the chance is exceedinglyunlikely.)If a mutation does occur in the jth cycle (i.e., among the 2j cells that result in this cycle, where one is a new mutant),the fraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:4   MutantCoverage.nbbased on one minus the probability that no mutation hits.  (Technically, this allows for the possibility that more thanone hit would occur at the exact same site in different cells in the same cell cycle, but the chance is exceedinglyunlikely.)If a mutation does occur in the jth cycle (i.e., among the 2j cells that result in this cycle, where one is a new mutant),the fraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:frac@j_D = 1 ë 2j;Thus, prob[j,m] gives us the probability distribution for the fraction, frac[j], of mutant cells in the source population(amounting to a number of mutant cells N1 ë 2j), with the probability of at least one mutant cell at a particular siteequalling:mutantprob@m_, N1_D = Sum@prob@j, mD, 8j, 1, cycles<D‚j=1Log@N1DLog@2D I1 - H1 - 2 mL2-1+jMFor the mutation rate of Zhu et al. and the population size estimated for BY, the probability of at least one mutant cellin the source population is:mutantprob@trymZHU, tryNBYD0.17409The expected number of mutant cells is the chance of a mutation occurring at cycle j and the number of cells that result(N1*frac[j]), summed over all cell divisions:Sum@prob@j, trymZHUD * tryNBY * frac@jD, 8j, 1, tryc<D3.39431Finally, we calculate the probability that one or more mutant cells will be placed in the flask, given that a fraction, f, ofthe N1 cells were sampled:probhit@m_, N1_, f_, c_D =1 - H1 - mutantprob@m, N1DL - SumAprob@j, mD * H1 - frac@jDLf*N1 , 8j, 1, c<EH*We calculate the probability of sampling some mutantcells as one minus the probability of sampling none,either because no mutations occur H1-mutantprob@m,N1DL or becausemutations occur in cycle c but are not sampled Hthe sumL*L‚j=1Log@N1DLog@2D I1 - H1 - 2 mL2-1+jM - ‚j=1c I1 - 2-jMf N1 I1 - H1 - 2 mL2-1+jMprobhit@trymZHU, tryNBY, tryf, trycD0.0473538Given L independent flasks (each started from a different colony), where each lineage is started with a fraction, f, of itsown source population,  the probability that at least one mutant cell will be sampled into at least one of the flaskpopulations would be:1 - H1 - probhit@m, N1, f, cDLL1 - 1 - ‚j=1Log@N1DLog@2D I1 - H1 - 2 mL2-1+jM + ‚j=1c I1 - 2-jMf N1 I1 - H1 - 2 mL2-1+jM LMutantCoverage.nb   51 - H1 - probhit@trymZHU, tryNBY, tryf, trycDLtryL0.384375The expected number of flasks with at least one mutant cells bearing a mutation at the focal site is then:tryL * probhit@trymZHU, tryNBY, tryf, trycDH*Expected number of flasks with mutations.*L0.473538If we multiply by the average ORF length, we would expect 656 mutations per ORF to have arisen across the 10 flasks:tryL * probhit@trymZHU, tryNBY, tryf, trycD * tryORF655.851Of these mutations per ORF, we expect each gene would have, on average, 32 premature stop codons somewhere inone of  the  flasks  (fraction of  non-sense mutations  in  a  gene estimated in  Gerstein  et  al.  (2015)’s  SupplementaryMathematica package, available on DRYAD, DOI: https://doi.org/10.5061/dryad.5gp25):tryL * probhit@trymZHU, tryNBY, tryf, trycD * tryORF * 0.048831111111111124`32.0259Similarly,  Each gene would have, on average, 483 AA changes somewhere in one of the flasks (fraction of non-synonymous mutations in a gene estimated in Gerstein et al. (2015)’s Supplementary Mathematica package, availableon DRYAD, DOI: https://doi.org/10.5061/dryad.5gp25):tryL * probhit@trymZHU, tryNBY, tryf, trycD * tryORF * 0.7363755555555553`482.952ü ConclusionThe above calculations inform us that each site within the genome is likely to have been hit by a nucleotide changingmutation (expected number of deep well hits = 0.35; expected number of flask hits = 0.47; expected number of totalhits = 0.82).  As a consequence, we are likely to have sampled ~888 non-synonymous or non-sense mutations for eachgene within the genome over the course of the experiment.Caveats:  Of course, the fact that a single cell of the right genotype is sampled into one of the wells doesn’t mean thatit will necessarily establish; it may die before dividing.  Plus, the above calculations used the average mutation rate;sites with lower mutation rates are less likely to have been sampled.  Furthermore, we only calculate the chance that asite mutates, not the chance of having sampled all three possible alternative nucleotides (which would require account-ing for differences in transition and transversion mutation rates).  Finally, the above calculations are only for SNPs andignore more complex mutations (indels, rearrangements, LOH, etc.)Nevertheless, the above calculations demonstrate that the acquisition experiments had a reasonable chance of explor-ing SNP mutations at most sites within the genome.Chance that a two-step mutation would ariseü Deep Well BoxesHere we modify the above to calculate the chance of observing two mutations within the same gene within a well,providing resistance even if only recessive mutations are available.  We consider two cases, either where the secondarymutation can occur anywhere within the same ORF (in the homologue) or where the secondary mutation is a loss-of-heterozygosity event (estimated to occur by mitotic recombination at a rate of ~0.8 µ 10-4; Mandegar & Otto, 2007,Proc Roy Soc B; only half of which is assumed to lead to the homozygous recessive mutant).Additional parameters:m2 = secondary mutation rate (assumed to be either m × ORF or 0.8 µ 10-4 ë 2 )N1 = population size at saturation in 150uL rich medium for growth in bioscreenf = fraction of population sampled to found a lineage (0.0667 = 10 ul/150 uL)L = # diploid deep wells (619: {191-13,286,155} in the three acquisition experiments in deep well boxes)ORF = average length of an ORF (1385 from Hurowitz, E. H., & Brown, P. O. 2003 Genome biology, 5(1),R2.)6   MutantCoverage.nbAdditional parameters:m2 = secondary mutation rate (assumed to be either m × ORF or 0.8 µ 10-4 ë 2 )N1 = population size at saturation in 150uL rich medium for growth in bioscreenf = fraction of population sampled to found a lineage (0.0667 = 10 ul/150 uL)L = # diploid deep wells (619: {191-13,286,155} in the three acquisition experiments in deep well boxes)ORF = average length of an ORF (1385 from Hurowitz, E. H., & Brown, P. O. 2003 Genome biology, 5(1),R2.)trymZHU = 1.67 µ 10-10;trymLYNCH = 3.3 µ 10-10;tryNBY = 7.03 * 107 * 0.15;H*Estimated population density per mL for BY and scaling to 150uL YPD.*LtryN303 = 7.42 * 107 * 0.15;H*Estimated population density per mL for W303 and scaling to 150uL SC.*Ltryf = 0.0667;tryL = 8191 - 13, 286, 155<; H*Number of diploid wells per acquisition experiment,excluding ones later found to have been haploid contaminated*LtryORF = 1385;trym2a = trymZHU * tryORF;H*Assuming that any other mutation in the same gene would inactivate,a conservative assumption.*Ltrym2b = 0.8 µ 10-4 ë 2; H*Assuming a secondary LOH event,half of which lead to the homozygous recessive mutant.*LNumber of cell cycles required to produce source population of N1 cells:cycles = Log@2, N1DLog@N1DLog@2DIn cases where we are performing numerical sums and require an integer number, we round down the number of cycles(rounding down is slightly conservative):tryc = Floor@cycles ê. N1 Ø tryNBYD23Total number of cell divisions involved (1 cell division from 1 Ø 2 cells, 2 cell divisions from 2 Ø 4 cells, etc):divisions = Sum@2^i, 8i, 0, cycles - 1<D-1 + N1The chance that NONE of these cell divisions produced a mutation at a particular site in a diploid (bearing 2m muta-tions per cell division across the two homologues) is:H1 - 2 mLdivisions;The distribution of mutant cell numbers in the source population is broad and very skewed (a “jackpot distribution”),and it is possible that the mutation hit early and generated a lot of mutant cells.  To account for this mutational distribu-tion, the probability that a mutation at one specific site occurs in the jth cell cycle (going from m = 2j-1  cells to 2jcells)  is:prob@j_, m_D = 1 - H1 - 2 mLm ê. m Ø 2j-1;based on one minus the probability that no mutation hits.  (Technically, this allows for the possibility that more thanone hit would occur at the exact same site in different cells in the same cell cycle, but the chance is exceedinglyunlikely.)If a mutation does occur in the jth cycle (i.e., among the 2j cells that result in this cycle, where one is a new mutant),the fraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:frac@j_D = 1 ë 2j;The number of remaining cell cycles is “cycles-j”, during which there will be 2Hcycles-jL - 1  divisions among thecells that already carry the first mutation (using the same logic used above to get “divisions”), so that the chance that amutation occurs and then bears a secondary mutation is:MutantCoverage.nb   7The number of remaining cell cycles is “cycles-j”, during which there will be 2Hcycles-jL - 1  divisions among thecells that already carry the first mutation (using the same logic used above to get “divisions”), so that the chance that amutation occurs and then bears a secondary mutation is:SumAprob@j, mD * m2 * I2Hcycles-jL - 1M, 8j, 1, cycles<E‚j=1Log@N1DLog@2D I-1 + 2-j N1M I1 - H1 - 2 mL2-1+jM m2For a secondary mutation in the same ORF:SumAprob@j, trymZHUD * trym2a * I2Htryc-jL - 1M, 8j, 1, tryc<E6.80428 µ 10-9For a secondary LOH event:SumAprob@j, trymZHUD * trym2b * I2Htryc-jL - 1M, 8j, 1, tryc<E1.17673 µ 10-6The above just calculates the chance that a two-step mutation is in the source population.  The chance that it will besampled requires that we first calculate the expected number of two-step mutant cells in the source population.The chance that a first mutation occurs in the jth cell cycle (going from m = 2j-1  cells to 2j   cells) and then a sec-ondary mutation occurs in the kth cell  cycle among the cells that bear the first  mutation (going from n = 2k-j-1mutant cells to 2k-j mutant cells) is:prob2@j_, k_, m_, m2_D = H1 - H1 - 2 mLmL H1 - H1 - m2LnL ê. m Ø 2j-1 ê. n Ø 2k-j-1;(this assumes that the chance of both mutations happening in the same cell division is negligible and assumes that onlymutations in the homologue can generate resistance).If a secondary mutation does occur in the kth cycle after the first one in the jth cycle (when there are 2k  cells), thefraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:frac2@j_, k_D = 1 ë 2k;Thus, prob2 gives us the probability distribution for the fraction, frac2, of secondary mutant cells in the source popula-tion (amounting to a number of mutant cells N1 ë 2k), with the probability of at least one two-step mutant cell at aparticular site equalling:mutantprob@m_, m2_, N1_D = Sum@prob2@j, k, m, m2D, 8j, 1, cycles - 1<, 8k, j + 1, cycles<D‚j=1-1+ Log@N1DLog@2D ‚k=1+jLog@N1DLog@2D I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kMFor a secondary mutation in the same ORF:mutantprob@trymZHU, trym2a, tryNBYD6.70832 µ 10-9For a secondary LOH event:mutantprob@trymZHU, trym2b, tryNBYD8.77605 µ 10-7Finally, we calculate the probability that one or more secondary mutant cells will be placed in the deep well box, giventhat a fraction, f, of the N1 cells were sampled:8   MutantCoverage.nbprobhit2@m_, m2_, N1_, f_, c_D = 1 - H1 - mutantprob@m, m2, N1DL -SumAprob2@j, k, m, m2D * H1 - frac2@j, kDLf*N1 , 8j, 1, c - 1<, 8k, j + 1, c<EH*We calculate the probability of sampling some two-step mutant cells as one minus the probability of sampling none,either because no two-step mutations occur H1-mutantprob@m,m2,N1DL or because two-step mutations occur but are not sampled Hthe sumL*L‚j=1-1+ Log@N1DLog@2D ‚k=1+jLog@N1DLog@2D I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kM -‚j=1-1+c ‚k=1+jc I1 - 2-kMf N1 I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kMFor a secondary mutation in the same ORF:probhit2@trymZHU, trym2a, tryNBY, tryf, trycD1.24781 µ 10-9For a secondary LOH event:probhit2@trymZHU, trym2b, tryNBY, tryf, trycD1.72446 µ 10-7Given L independent deep wells (each started from a different colony), where each lineage is started with a fraction, f,of its own source population,  the probability that at least one secondary mutant cell will be sampled into at least one ofthe deep well populations would be:1 - H1 - probhit2@m, m2, N1, f, cDLL1 - 1 - ‚j=1-1+ Log@N1DLog@2D ‚k=1+jLog@N1DLog@2D I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kM +‚j=1-1+c ‚k=1+jc I1 - 2-kMf N1 I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kM LFor a secondary mutation in the same ORF:1 - H1 - probhit2@trymZHU, trym2a, tryNBY, tryf, trycDLTotal@tryLD7.72393 µ 10-7For a secondary LOH event:1 - H1 - probhit2@trymZHU, trym2b, tryNBY, tryf, trycDLTotal@tryLD0.000106739ü FlasksHere we modify the above to calculate the chance of observing two mutations within the same gene within a flask,providing resistance even if only recessive mutations are available.  We consider two cases, either where the secondarymutation can occur anywhere within the same ORF or where the secondary mutation is a loss-of-heterozygosity event(estimated to occur by mitotic recombination at a rate of ~0.8 µ 10-4; Mandegar & Otto, 2007, Proc Roy Soc B; onlyhalf of which is assumed to lead to the homozygous recessive mutant).Additional parameters:m2 = secondary mutation rate (assumed to be either m × ORF or 0.8 µ 10-4 ë 2)N1 = population size at saturation in 10mL YPDf = fraction of population sampled to found a lineage (0.1 = 1 mL/10 mL)L = # diploid flasks in expt (10)ORF = average length of an ORF (1385 from Hurowitz, E. H., & Brown, P. O. 2003 Genome biology, 5(1),R2.)MutantCoverage.nb   9Additional parameters:m2 = secondary mutation rate (assumed to be either m × ORF or 0.8 µ 10-4 ë 2)N1 = population size at saturation in 10mL YPDf = fraction of population sampled to found a lineage (0.1 = 1 mL/10 mL)L = # diploid flasks in expt (10)ORF = average length of an ORF (1385 from Hurowitz, E. H., & Brown, P. O. 2003 Genome biology, 5(1),R2.)trymZHU = 1.67 µ 10-10;tryNBY = 7.03 * 107 * 10;H*Using estimated population density per mL and scaling up to 10mL*Ltryf = 0.1; H*1mL into 10mL*LtryL = 10;tryORF = 1385;trym2a = trymZHU * tryORF;H*Assuming that any other mutation in the same gene would inactivate,a conservative assumption.*Ltrym2b = 0.8 µ 10-4 ë 2; H*Assuming a secondary LOH event,half of which lead to the homozygous recessive mutant.*LNumber of cell cycles required to produce source population of N1 cells:cycles = Log@2, N1DLog@N1DLog@2DIn cases where we are performing numerical sums and require an integer number, we round down the number of cycles(rounding down is slightly conservative):tryc = Floor@cycles ê. N1 Ø tryNBYD29Total number of cell divisions involved (1 cell division from 1 Ø 2 cells, 2 cell divisions from 2 Ø 4 cells, etc):divisions = Sum@2^i, 8i, 0, cycles - 1<D-1 + N1The chance that NONE of these cell divisions produced a mutation at a particular site in a diploid (bearing 2m muta-tions per cell division across the two homologues) is:H1 - 2 mLdivisions;The distribution of mutant cell numbers in the source population is broad and very skewed (a “jackpot distribution”),and it is possible that the mutation hit early and generated a lot of mutant cells.  To account for this mutational distribu-tion, the probability that a mutation at one specific site occurs in the jth cell cycle (going from m = 2j-1  cells to 2jcells)  is:prob@j_, m_D = 1 - H1 - 2 mLm ê. m Ø 2j-1;based on one minus the probability that no mutation hits.  (Technically, this allows for the possibility that more thanone hit would occur at the exact same site in different cells in the same cell cycle, but the chance is exceedinglyunlikely.)If a mutation does occur in the jth cycle (i.e., among the 2j cells that result in this cycle, where one is a new mutant),the fraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:frac@j_D = 1 ë 2j;The number of remaining cell cycles is “cycles-j”, during which there will be 2Hcycles-jL - 1  divisions among thecells that already carry the first mutation (using the same logic used above to get “divisions”), so that the chance that amutation occurs and then bears a secondary mutation is:10   MutantCoverage.nbSumAprob@j, mD * m2 * I2Hcycles-jL - 1M, 8j, 1, cycles<E‚j=1Log@N1DLog@2D I-1 + 2-j N1M I1 - H1 - 2 mL2-1+jM m2For a secondary mutation in the same ORF:SumAprob@j, trymZHUD * trym2a * I2Htryc-jL - 1M, 8j, 1, tryc<E5.59293 µ 10-7For a secondary LOH event:SumAprob@j, trymZHUD * trym2b * I2Htryc-jL - 1M, 8j, 1, tryc<E0.0000967238The above just calculates the chance that a two-step mutation is in the source population.  The chance that it will besampled requires that we first calculate the expected number of two-step mutant cells in the source population.The chance that a first mutation occurs in the jth cell cycle (going from m = 2j-1  cells to 2j   cells) and then a sec-ondary mutation occurs in the kth cell cycle among the cells that bear the first mutation (going from n = 2k-j-1 cellsto 2k-j mutant cells) is:prob2@j_, k_, m_, m2_D = H1 - H1 - 2 mLmL H1 - H1 - m2LnL ê. m Ø 2j-1 ê. n Ø 2k-j-1;(this assumes that the chance of both mutations happening in the same cell division is negligible and assumes that onlymutations in the homologue can generate resistance).If a secondary mutation does occur in the kth cycle after the first one in the jth cycle (when there are 2k  cells), thefraction of the source population that will be mutant (assumed to be unaffected by selection prior to placement innystatin) is:frac2@j_, k_D = 1 ë 2k;Thus, prob2 gives us the probability distribution for the fraction, frac2, of secondary mutant cells in the source popula-tion (amounting to a number of mutant cells N1 ë 2k), with the probability of at least one two-step mutant cell at aparticular site equalling:mutantprob@m_, m2_, N1_D = Sum@prob2@j, k, m, m2D, 8j, 1, cycles - 1<, 8k, j + 1, cycles<D‚j=1-1+ Log@N1DLog@2D ‚k=1+jLog@N1DLog@2D I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kMFor a secondary mutation in the same ORF:mutantprob@trymZHU, trym2a, tryNBYD4.76763 µ 10-7For a secondary LOH event:mutantprob@trymZHU, trym2b, tryNBYD0.0000562647Finally, we calculate the probability that one or more secondary mutant cells will be placed in a flask, given that afraction, f, of the N1 cells were sampled:MutantCoverage.nb   11probhit2@m_, m2_, N1_, f_, c_D = 1 - H1 - mutantprob@m, m2, N1DL -SumAprob2@j, k, m, m2D * H1 - frac2@j, kDLf*N1 , 8j, 1, c - 1<, 8k, j + 1, c<EH*We calculate the probability of sampling some two-step mutant cells as one minus the probability of sampling none,either because no two-step mutations occur H1-mutantprob@m,m2,N1DL or because two-step mutations occur but are not sampled Hthe sumL*L‚j=1-1+ Log@N1DLog@2D ‚k=1+jLog@N1DLog@2D I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kM -‚j=1-1+c ‚k=1+jc I1 - 2-kMf N1 I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kMFor a secondary mutation in the same ORF:probhit2@trymZHU, trym2a, tryNBY, tryf, trycD1.27163 µ 10-7For a secondary LOH event:probhit2@trymZHU, trym2b, tryNBY, tryf, trycD0.0000151389Given L independent deep wells (each started from a different colony), where each lineage is started with a fraction, f,of its own source population,  the probability that at least one secondary mutant cell will be sampled into at least one ofthe deep well populations would be:1 - H1 - probhit2@m, m2, N1, f, cDLL1 - 1 - ‚j=1-1+ Log@N1DLog@2D ‚k=1+jLog@N1DLog@2D I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kM +‚j=1-1+c ‚k=1+jc I1 - 2-kMf N1 I1 - H1 - 2 mL2-1+jM I1 - H1 - m2L2-1-j+kM LFor a secondary mutation in the same ORF:1 - H1 - probhit2@trymZHU, trym2a, tryNBY, tryf, trycDLtryL1.27163 µ 10-6For a secondary LOH event:1 - H1 - probhit2@trymZHU, trym2b, tryNBY, tryf, trycDLtryL0.000151379ü ConclusionAllowing secondary mutations to occur among the diploid cells that bear a first mutation, there is <0.0003 chance thata secondary mutant cell will be sampled in any of the deep wells or flasks in the experiment, assuming that the sec-ondary mutation rate or LOH rate is <~10-4.  We consider this to be a conservative estimate of the secondary muta-tion rate because we have assumed that any mutation in the same ORF on the homologue would generate resistance, aswould mitotic recombination whenever it creates a homozygous mutant.12   MutantCoverage.nbC.3. Supporting TableC.3 Supporting TableTable C.1: Results of ANOVA and post-hoc Tukey tests comparing mean number of days until growthbetween the different strain types in each experiment. For the Tukey tests, estimates are reported with p-values in parentheses. Estimates are negative when the second group has a higher mean. Significant p-valuesare in bold.Acquisition Experiment 1 Acquisition Experiment 2 Acquisition Experiment 3ANOVAF 14.45 61.25 97.55df 2, 206 2, 161 2, 229p-value 1.35 ⇥ 106 < 1015 < 1015Pairwise comparisons (Tukey)MATa - diploid -2.26 (< .0001) -2.86 (< .0001) -1.87 (< .0001)MAT↵ - diploid -1.13 (0.0043) -2.54 (< .0001) -0.27 (0.29)MATa - MAT↵ -1.13 (0.023) -0.31 (0.75) -1.60 (< .0001)155C.4. Supporting FiguresC.4 Supporting Figures0 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))MATa0 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))MATα0 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 10 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 20 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 30 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 40 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 50 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 60 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 70 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 80 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 90 12 24 36 48 60 720.00.51.01.5d.3[, 1]seq(0, 1.9, length = length(d.3[, 1]))Diploid 10Optical DensityTime (h)Figure C.1: Growth curves of all populations from larger volume flasks. All populations grew in the originalacquisition experiment by Day 10, but only the haploid populations show reliable growth when re-tested inYPDnystatin4. No diploid populations grew over the 72 hour assay.156C.4. Supporting Figures●●●●●●●●●●●●●●● ● ●● ●●● ●●● ●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●1 2 3 4 5 6 7 8 9 10 11 12Growth DayOD720.00.51.01.5Acquisition Experiment 1●●● ●●●●●●● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●1 2 3 4 5 6 7 8 9 10Growth DayOD72Acquisition Experiment 2●●●●●●●●● ●●●●●●●●●●●●●●●●●●● ● ●● ●●● ●●●●●●●●● ●●● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●●● ●● ●●●●●●●●●●●●●●●●●●●●●●●● ●●●● ●●●● ●●●●●●●●●●●●●●●●●●●1 2 3 4 5 6 7Growth DayOD72MATalphaMATaDiploidAcquisition Experiment 3OD72Growth DayFigure C.2: Plot of mean OD72 when populations are re-tested in nystatin versus the number of days untilgrowth was observed in the original acquisition experiments in deep well boxes. Black dashed lines indicatethe range of cutoff values used for judging a potential mutant strain when growth was re-tested. Solidlines indicate the fitted regression lines from linear regressions (black: all populations considered together,coloured: model run using only the populations of that type). Note that the x-axis changes depending on thelength of the acquisition experiment.157

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0362398/manifest

Comment

Related Items